Understanding Variation in Water Quality using a Riverscape Perspective A thesis submitted in partial fulfilment of the requirements for the Degree of Master of Science in Zoology in the University of Canterbury New Zealand By Hannah M. Franklin University of Canterbury 2010
147
Embed
Understanding Variation in Water Quality using a Riverscape … · 2016-05-26 · Quality using a Riverscape Perspective . A thesis submitted in partial fulfilment of the requirements
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Understanding Variation in Water
Quality using a Riverscape Perspective
A thesis submitted in partial fulfilment of the
requirements for the Degree of
Master of Science in Zoology
in the University of Canterbury
New Zealand
By
Hannah M. Franklin
University of Canterbury
2010
Table of Contents
Table of Contents Acknowledgements..........................................................................................................................1
ACKNOWLEDGMENTS I would like to thank Professor Angus McIntosh (University of Canterbury) for his ongoing
support and supervision through the completion of my studies. From day one his enthusiasm and
energy guided me throughout the development of my project. The prompt response in answering
queries and reviewing written material was much appreciated. I have enjoyed working with and
getting to know Angus. I would also like to thank Associate Professor Jon Harding (University
of Canterbury) for his advice along the way. I have learnt so much from both Angus and Jon, and
have a great deal of respect for them in their roles heading such an active and enthusiastic group.
Many thanks also go to those that helped edit drafts in particular Jon O’Brien and Rebecca
Campbell (University of Canterbury).
I have thoroughly enjoyed being part of the Freshwater Ecology Research Group (FERG)
and have appreciated their ongoing support in critiquing my research methods and reports. I have
really enjoyed getting to know the group, wish you all the best with your respective projects and
hope to meet again in our future careers.
Finally, thank you also to my family, flatmates and friends for their support and putting
up with me throughout this project. In particular, I cannot express how grateful I am for the help
and encouragement of Sarah Dawson, Anna Henderson, Anne Braithwaite and Moritz Lasse, I
will not forget this.
2
ABSTRACT With the increasing degradation of rivers worldwide, an understanding of spatial and temporal
patterns in freshwater quality is important. Water quality is highly variable in space and time, yet
this is largely overlooked at the scale of stream catchments. I employed a landscape ecology
approach to examine the spatial patterning of water quality in complex, impacted stream
networks on the Canterbury Plains of the South Island of New Zealand, with the goal of
understanding how land-use effects proliferate through stream systems.
In particular, I used “snapshot” sampling events in conjunction with spatial modelling
and longitudinal profiles to investigate the ways in which spatial and environmental factors
influence the variability of water quality in stream networks. Spatial eigenfunction analyses
showed that distance measures, which took into account variable connectivity by flow and
distance along the stream between sites, explained more spatial variance in water quality than
traditional distance metrics. Small upstream reaches were more spatially and temporally variable
than main stems (under summer base-flow conditions). The extent of spatial variation in water
quality differed between stream networks, potentially depending on linkages to groundwater and
the surrounding landscape. My results indicated that the water quality of headwater streams can
have a disproportionate influence over water quality throughout an entire network.
I investigated spatio-temporal patterns in water quality more intensively in one stream
network, the Cam River, in which I found consistent spatial pattern through time. The relative
balance between nutrient inputs (pollution and groundwater) and in-stream conditions influenced
the spatial pattern of water quality, as well as that of several ecosystem processes which I
measured simultaneously. The spatially intensive and explicit approach has allowed
identification of key factors controlling water quality and ecosystem processes throughout the
Cam River. This research highlights the importance of taking a spatially explicit approach when
studying stream water quality and that such an approach could be insightful and will contribute
to solving current stream management problems.
Chapter One 3
1 CHAPTER ONE - UNDERSTANDING VARIATION IN
WATER QUALITY USING A RIVERSCAPE PERSPECTIVE
.
An example of the heterogeneity of stream conditions within a single, relatively small catchment,
the L2 River on the Canterbury Plains.
Chapter One 4
Early stream ecologists identified that streams are the product of their catchments (Hynes 1975).
However, that streams are connected dendritic networks, has only recently been included
explicitly in stream research (Stewart-Koster et al. 2007, Brown and Swan 2010). The flowing,
branched nature of streams means that water quality within a stream network is intrinsically
spatially correlated (Tu and Xia 2008). However, development of methods to describe spatial
patterns within stream systems has lagged behind those in terrestrial ecosystems (Blanchet et al.
2008b). In this thesis I aimed to show, not only that taking a spatially explicit approach was
necessary when studying stream water quality, but also that such an approach could be insightful and
will contribute to solving current stream management problems.
1.1 DEVELOPMENT OF THE RIVERSCAPE APPROACH
Spatial heterogeneity is the most salient feature of landscapes (Wu and Hobbs 2007).
Understanding the patterns, causes and consequences of spatial heterogeneity for ecosystem
function is a key research topic in many branches of ecology (Dutilleul and Legendre 1993,
Turner and Cardille 2007). Landscape ecology theory holds that heterogeneous spatial patterns
matter, as they set the context for organic processes such as fluxes of organisms, matter and
energy. Landscape ecology sees various scales, and the interplay between these, as important
(Fausch et al. 2002, Talley 2007). This approach has traditionally been carried out in spite of the
statistical noise from spatial variation. However over the last decade, landscape ecology has
made great advances in theory and practice (Wu and Hobbs 2007). What has been described as
the “emerging science of scale” (Wiens 2002) has led to the recognition that studying spatial
structure is both a requirement for ecologists and a challenge. This is a new paradigm (Legendre
1993). Some idea of the emerging importance of space can be gleaned from the bibliometric
approach. For example, a clear increase in the number of publications using the words “spatial”
and “autocorrelation” in the title or abstract can be seen after 1990 (Rangel et al. 2006). Many
Chapter One 5
new statistical packages are available which incorporate new statistical methodology developed
to treat space more explicitly in analyses (Rangel et al. 2010).
Streams are an integrating element in complex landscapes incorporating much spatial and
temporal heterogeneity (Thompson and Lake 2009, Brown and Swan 2010). Many landscape
theories have been applied, tested and developed for streams (Thompson and Lake 2009).
Drawing on these ideas and incorporating them into freshwater research will allow a more
accurate picture of ecological processes to be gained (Fausch et al. 2002). Early models of
stream ecosystems, such as the River Continuum Concept (RCC), incorporated landscape
elements only in a basic way. The RCC considered physical, chemical and biological attributes
of a stream to be changing along longitudinal gradients from headwaters to lower reaches
(Vannote et al. 1980). Gradually the consideration of landscape ideas has moved into stream
ecology. The serial discontinuity concept made some headway towards a concept of rivers as
interrupted continua (resulting from human alteration). This concept also included the idea that a
change in any physical or biological parameter may be felt in an upstream or downstream
direction of a flow obstruction (Ward and Stanford, 1995). The effects of disturbance from water
abstraction, land-use changes and point source discharges can similarly act in both directions
(Pringle, 1997). Recently a multidimensional perspective of rivers has been developed. Vertical
and lateral linkage between river channels and the adjacent subsystems are important; hyporheic,
parafluvial and riparian, have now been included in various models (Fisher et al. 1998, Junk and
Wantzen, 2004).
In recent years stream ecology has emphasised links between streams and landscapes.
Terrestrial frameworks such as patches, fragmentation and hierarchies have moved into the
freshwater realm (Boulton et al. 1997, Torgersen et al. 1999, Talley 2007). In the previously
mentioned concepts it is recognised that rivers are influenced by the landscape through which
they flow. However, the spatial scale under which these concepts are often applied is small.
Although stream systems have long been recognised as having hierarchical structure, most
Chapter One 6
studies still take place at the reach scale, which itself is poorly defined. However, stream
management must occur at larger scales, within catchments, or sub-catchments (Lowe et al.
2006, Thompson and Lake 2009).
Various attempts have been made to embed these ideas in a larger scale approach
(Melbourne and Chesson 2006) and studies have been conducted incorporating whole
catchments, referred to as the “riverscape” perspective (Allan 2004). From this standpoint, rivers
are investigated as whole ecosystems, strongly influenced by their surrounding landscapes, and
at multiple scales (Wiens 2002, Allan 2004). The wider landform through which a stream flows
varies in terms of geology, hydrology, vegetation, topography and climate. Landscape factors
exert variable influences over in-stream parameters, processes and biota throughout systems. The
importance of considering how the configuration of tributaries and main channels within a
stream system relate to the wider landscape has been recognised (Townsend et al. 2003, Allan
2004). However, many studies do not account for the reality of the connected configuration of
stream networks. This is of particular consequence in riverine networks, which consist of linearly
arranged, hierarchical, branching habitats (Fagan 2002, Brown and Swan 2010). Fagan (2002)
predicted that because of their hierarchical nature, the within-network flux of organisms would
be greater through mainstems relative to headwaters, as these include movement of organisms
and matter from and between branches. The dispersal rate of organisms and matter are
disproportionately important in riverine systems due to the asymmetric movement of energy and
matter from headwaters to stems. Consequently forces structuring communities may vary
drastically in different parts of a network (Brown and Swan 2010). Researchers who study water
quality in stream networks encounter issues that go beyond those already mentioned. Dendritic
landscapes have a striking mismatch between the geometry of dispersal, “as the trout swims”,
and the geometry of disturbance, “as the crow flies” (Fagan 2002). Moreover, it has been
suggested that stream distance is a more appropriate representation than Euclidean distance,
Chapter One 7
when the travel path between two locations is restricted to the stream (Fagan 2002, Gardner et al.
2003, Lyon et al. 2008).
The source of water in streams is spatially variable, with the intermixing of ground and
surface waters occurring at different spatial scales and dimensions (Gregory et al. 1991, Cruezé
des Châtelliers et al. 1994, Dahm et al. 1998). Landscape elements, such as tributaries, channel
gradient, width to depth ratio, sinuosity, relic channels, geolithology, channel morphology and
stream bed composition, add further complexity to the exchange of surface and ground waters
(Dahm et al. 1998, Wayland et al. 2003). These may cause variation in the intermixing of surface
and ground waters, which in turn may cause spatially variable water quality. Artificial
channelisation and straightening speed the flow of surface waters and minimise connectivity to
ground water and riparian systems. Such reaches are thought to function more like a pipe with
little nutrient and sediment retention (Dahm et al. 1998). Alternatively, tributaries and aggrading
reaches have increased interaction with ground water (Gregory et al. 1991). Spatially
heterogeneous geomorphic processes are discontinuous due to variation in grain size, tributary
inputs and vertical segregation through variable aggradations and degradation (Cruezé des
Châtelliers et al. 1994). However, most water quality research is focused at either the smallest or
largest scales (Dahm et al. 1998, Lyon et al. 2008). Fine scale studies include those of the path of
water into the stream at single location (Peterjohn and Correll 1984, Sheridan et al. 1999) and
reach-scale research on nutrient cycling (Niyogi et al. 2004). At the other end of the spectrum,
catchment water quality is usually monitored on a regular basis at a small number of locations in
a catchment, generally focused at the catchment outlet. This integrates the effect of all the point
and non-point source processes occurring throughout the catchment. However, effective
catchment management requires data which identifies major sources and processes which
operate between these scales (Grayson et al. 1997, Salvia et al. 1999).
Classic theory on the processes influencing catchment water quality comes from two
sources, catchment ecology and nutrient spiralling. Research in the Hubbard Brook catchment
Chapter One 8
involved nutrient budgets of precipitation inputs and stream outputs; this typified the catchment
approach (Lowe and Likens 2005). In the catchment approach, riparian and hyporheic zones held
little importance, but chemical changes of water moving through these zones suggest they play
an important role in nutrient transformation (Peterjohn and Correll 1984, Valett et al. 1996).
From the nutrient spiralling perspective, nutrients were studied in a production sense, where the
focus was on budgets within the wetted stream only (Newbold et al. 1982, Grimm 1987).
Nutrient cycling processes have been shown to change the concentration of nutrients in the water
column, but the focus is turning to where these processes occur and their relative importance
(Finlay et al. 2011). Merging these perspectives will increase our understanding of catchment
nutrient processes (Dahm et al. 1998). To be meaningful, studies considering the outflow of
solutes from a drainage basin must also consider the aggregation of upstream solute responses
and channel routing (Walling and Webb 1980). Typically studies lump the whole drainage basin
together in analysis, not considering basin size or heterogeneity. More particularly they mostly
ignore the fact the upstream drainage basins may be additive in their downstream effect, and that
the course of this effect is determined by channel routing (Walling and Webb 1980). Knowledge
of hydrological flow paths is critical in linking the hierarchy of stream networks (Dahm et al.
1998). Thus, in Chapter Two, I examined spatial patterns in water quality within five complex
dendritic stream systems with a focus on factors influencing the level of spatial variability
observed in each system.
1.2 BASE FLOW WATER QUALITY SAMPLING
Besides spatial autocorrelation, variation in flow levels, not only during flood events; influences
the concentration of many nutrients and levels of physico-chemical parameters (Meybeck et al.
1999, Scarsbrook 2002, Wayland et al. 2003). The water chemistry of rivers during base flow is
therefore more likely to represent the catchment geology and land-use than during flood flows or
droughts (Wayland et al. 2003). Thus, base flow sampling avoids the bias and influence of rivers
Chapter One 9
being sampled in particular states of flow, which would only further complicate analyses (Biggs
et al. 1990). Pionke et al. (1996) define base flow as the period between storms, when the
hydrograph is in the later stages of the recession limb, usually, more than two days post storm
flow peak.
Early base flow surveys were conducted from the perspective that in low flow, when
most flow comes from subsurface water stored in soil and rock, one would expect that water
quality will vary according to the geology within small catchments (Walling and Webb 1975). It
has been suggested that under base flow conditions, surface runoff is limited, so water is heavily
influenced by point sources and ground water, thus base flow conditions are a good time to
identify these effects (Eyre and Pepperell 1999). On the other hand, the continual presence of
base flow means that water chemistry in this period represents an integrated signal of climate,
geology, historical and present land-use throughout the watershed and should help link water
quality to land-use distributions (Wayland et al. 2003). The mean transit time of water (thus,
contact with the landscape) is also highest at low flow, so under these conditions, it is thought
that water quality will be tightly coupled with land-use (Lyon et al. 2008).
Base-flow conditions can exist for long periods, for example 10-11 months per year in
New Zealand (Biggs et al. 1990), so water quality at these flow levels is an important constraint
on the health of in-stream biological communities (Grayson et al. 1997, Lyon et al. 2008) and is
critical for water allocation (Biggs et al. 1990). However, caution should be used in
interpretation of data from only one sampling occasion, as base flow water chemistry may be
temporally variable in some watersheds (Clow et al. 1996) yet stable in others (Pionke et al.
1999). As a large portion of nitrates, phosphates and sediments are exported during floods, one
cannot scale-up to yearly catchment exports using base flow measurements (Dahm et al. 1998,
Eyre and Pepperell 1999, McKee et al. 2001).
Chapter One 10
1.3 THE “SNAPSHOT” APPROACH
Heterogeneity in both landscape and water quality exists within single catchments, even those
that are small (Prowse 1984) or in pristine condition (Clow et al. 1996, Finlay et al. 2011). The
consideration of sites across a network is important for improving understanding of the fate and
transport of nutrients, as well as for the identification of causal relationships (Tu and Xia 2008,
Finlay et al. 2011). A spatially intensive approach to water quality monitoring, which involves
the collection of water quality data from a large number of sample sites over a short period of
time, will achieve the best understanding of the processes occuring (Eyre and Pepperell 1999).
Walling and Webb (1975) were one of the first to publish a spatially intensive water quality
study. This involved relating conductivity within a catchment to geology and land-use. The
spatially intensive approach was not used significantly, however, until the 1990’s. It has been
redefined as “snapshot” or “synoptic” sampling (Grayson et al. 1997, Lyon et al. 2008
respectively).
The “philosophy” of the snapshot methodology is to sample a river at every confluence
point and discharge point at an instance in time (Walling and Webb 1975, Grayson et al. 1997).
The instant can be considered a period in which all parts of the river are at constant flow. As
most water quality parameters vary with discharge and floods, sampling during changing flow
would contain discharge-related variation, making spatial analysis difficult (Grayson et al. 1997).
Sampling in this way provides insight into the biogeochemical behaviour throughout a stream
network at low flow conditions and is in tune with emergent paradigms relating to spatial and
network perspectives in fresh water ecology (Fausch et al. 2002, Lyon et al. 2008). Lyon et al.
(2008) described a synoptic sampling campaign, with 100 or more sample locations, as still
spatially sparse compared to the heterogeneity found in natural stream systems. However,
published studies of snapshot sampling campaigns range between 50 sites sampled over 12 hours
(Salvia et al. 1999) to 108 in 3 days (Eyre and Pepperell 1999), with the most intensive being
over the longest time frame; over 500 in 12 days (Walling and Webb 1975). An exceptional
Chapter One 11
study is that of Dent and Grimm (1999), who collected around 40 samples almost simultaneously
along a 10km stretch.
The major advantages of the snapshot method relate to the bulk of additional information
that can be gained relatively quickly, compared with repeated data collection at catchment
outflows. This method has enabled insights into system behaviour, including quantification of
unknown point discharges, identification of key in-stream sources of suspended material and the
extent to which biological activity (phytoplankton growth) affects water quality (Grayson et al.
1997). The snapshot methodology also allows the mass balance approach to be used. Using this
approach allows all point-source total loads (tributary and drain inputs in this case) to be
subtracted from the total loads at main stem sites in a longitudinal manner (Grayson et al. 1997,
Salvia et al. 1999, Behrendt and Opitz 1999). In this way, snapshot sampling has been used as an
independent check of licensed discharges (Grayson et al. 1997). The large data sets provide
opportunities for cross-correlation with parameters that change across catchments. They have
allowed identification of geology, land-use and point source contributions to water quality, and
management action against point source contribution to be taken (Eyre and Pepperell 1999).
Snapshot sampling is often used to complement multiple measurements taken through time at
one site (Grayson et al. 1997). The regime sampled must be representative of base flow
conditions, as sampling in this way during a flood event would not meet the snapshot criteria
(Eyre and Pepperell 1999). Most studies which took a snapshot approach in sampling water
quality concluded that this method has been under-utilised and that it allowed valuable insight
into the properties of the system studied.
Repeated snapshot sampling events within a stream system have the potential to help
explain the level of temporal variability in stream water quality and have been associated with
flow, as well as changes in the spatial pattern of water quality through time (Clow et al. 1996,
Scarsbrook 2002, Wayland et al. 2003). The way streams interact with landscape factors through
time can vary, and the source of water may change (Dahm et al. 1998, Wayland et al. 2003). This
Chapter One 12
may result in temporal changes in spatial water quality patterns, yet only a handful of snapshot
studies have been conducted on a repeated basis. The logistical difficulties and cost involved in
snapshot sampling mean such studies have repeated the sampling event only three times across
variable time frames and successional stages of the hydrograph (Clow et al. 1996, Dent and
Grimm 1999, Wayland et al. 2003,). McKee et al. (2001) used a combined approach, sampling
using snapshot methodology six times at up to 79 sites across all seasons, routinely once a month
at subcatchment outlets and also event sampling up to six times a day during high flow events.
This sampling regime allowed patterns in water quality, longitudinally between sub-catchments
and through time, to be identified in this system (McKee et al. 2001). However, long term spatial
studies are rare. Exceptions are Prowse et al. (1984), who sampled 25 points many times over
three years, and a long-term monitoring project on the Seine River in France, where 236 stations
were spread across orders 1-8 and were sampled over a 30 year period (Meybeck et al. 1999).
Spatial changes in the physico-chemical composition of stream water through time may indicate
that the influence of various landscape and in-stream factors controlling water quality has
changed.
Short term variation in antecedent moisture, soil-water, microbial and dilution processes
will also vary (Pionke et al. 1999 and 1996). Thus, even base-flow water chemistry may be
temporally variable in some watersheds (Clow et al. 1996). Short term variations in water quality
can have flow-on effects in governing fish and insect assemblages and population distributions
(Stewart-Koster et al. 2007, Brown and Swan 2010). Such variation may be disproportionately
important under base-flow conditions, which can persist for much of the year. In light of the lack
of consensus in literature surrounding the role of slight variations in flow in controlling spatial
patterns in stream water chemistry through time, in Chapter Three I investigated spatio-temporal
variation in base-flow water quality in one of the five stream networks that I studied in Chapter
Two, the Cam River. In reducing the scale of study to a single catchment, I was able to sample in
Chapter One 13
a more spatially intensive and repeated manner, by undertaking multiple snapshot sampling
events.
Stream ecosystem processes, such as primary productivity and the turnover of organic
matter respond to the range of conditions experienced through time, in this way integrating some
of the temporal water quality variability often experienced (Biggs and Kilroy 2004). Such
processes are useful as a surrogate for water quality measurements, which are more expensive
(Bott et al. 1985). Ecosystem processes have direct impacts on consumers, the invertebrates and
fish which inhabit streams; and thus are of vital importance to sustaining aquatic biodiversity
(Winterbourn 2004). When measured through time (as bioaccumulation or breakdown rates) at
multiple points in a stream network, ecosystem processes may provide a “snapshot” of vigour
and resilience of biotic communities throughout.
The drivers and stressors of ecosystem processes have been studied at the large scale,
comparing processes across biomes (Bott et al. 1985), regions (Mullholand et al. 1987, Findlay
and Sinsabaugh 2006), stream types (Biggs and Close 1989) and between catchments (Quinn et
al. 1997). Comparisons to reference sites (Carlisle and Clements 2005, Ferreira et al. 2006) and
artificial enrichments (Gulis and Suberkropp 2003) are also common in stream productivity
research. Worldwide, studies of within-catchment variation in ecosystem processes are rare, and
the few I have found took place in relatively pristine catchments (Clow et al. 1996, Finlay et al.
2011). The lack of research into (and consequently knowledge of) within-catchment variation in
ecosystem processes may come from the fact that we know a lot about a few headwaters – such
as Hubbard Brook, yet processes in these systems cannot be applied to all (Bishop et al. 2008).
The study of degraded catchments, such as the Cam River, provides the opportunity to identify a
variety of relationships, which may be applicable to other complex, impacted systems. In
Chapter Four of this thesis I studied spatial variation in various ecosystem processes in the Cam
River, (which I had measured simultaneously with the snapshot water quality sampling events).
In this Chapter, I used spatial-environmental variance partition methods (first developed for
Chapter One 14
community ecology), in a novel way, to tease apart influences of anthropogenic impacts and
spatial location on variation in ecosystem processes within the Cam River system.
This thesis is structured as a series of stand alone papers intended for development
towards publication in international scientific journals. This means there is necessarily some
repetition, particularly in the Methods sections, but this is an effective and efficient way to
present the multiple aspects of this work. All analyses and writing are primarily my own, with
contributions of co-authors listed in the acknowledgements. Throughout the thesis, chapters are
referenced by chapter number as they appear in this thesis. Figures and tables are numbered
within each chapter, however the complete reference list is provided for all chapters at the end of
the thesis.
Chapter Two 15
2 CHAPTER TWO - SYSTEM-SPECIFIC SPATIAL
VARIABILITY OF STREAM WATER PHYSICOCHEMISTRY
The myriad of spring, lowland and foot-hills fed channels, as well as water-races and irrigation
canals, that cross the Canterbury Plains, forming many complex stream systems (as defined by the
River Environment Classification (REC, Snelder et al. 2005) (image from Google Inc. 2009, combined
with the REC in ArcMap 9.2).
Chapter Two 16
2.1 INTRODUCTION
Most of New Zealand was originally covered in forest or indigenous grasslands , yet since human
colonisation nearly two thirds of the forest cover has been removed. Conversion of this land to
agriculture and increased urbanisation have led to increases in nutrient levels (Houlahan and Findlay
2004, Simon et al. 2007), sedimentation (Suttle et al. 2004; Matthaei et al. 2006), microbial
contamination, and temperature (Larned et al. 2004) in freshwater streams, lakes, reservoirs and
wetlands and has led to the degradation of New Zealand's inland waters (Parkyn et al. 2003). These
influences are linked to the large-scale application of nutrient-rich fertiliser and livestock effluent.
The associated loss of riparian vegetation has contributed to these impacts and caused changes in
water quality by increasing the connectivity between the stream and the surrounding landscape
(Tabacchi et al. 1998; Parkyn et al. 2003). Animals crossing streams and stock pugging have also
had a deleterious effect on bank structure, decreasing nutrient retention and increasing sedimentation
(Houlahan and Findlay 2004; Collins et al. 2007). These physical changes can impact on the biotic
community and ecosystem processes of streams (Kronvang et al. 2005) and can result in the
degradation of water quality in streams and rivers.
Lowland streams on the Canterbury Plains exist in a landscape that has become a complex
mosaic with urban sprawl and changes in farming practices (Winterbourn 2008). Lowland streams
are also complex in terms of dendricity and hydrologic connectivity due to inputs from springs and
connectivity with stock and irrigation raceways. This is especially true for first order streams, which
make up a majority of the stream network. Small streams have been proposed as instrumental in
conditioning the water for export, dampening flood waters, cycling nutrients (Bernhardt et al. 2003)
and buffering pollutants (Klaminder et al. 2006); and also for habitat and refugia (Lacey et al. 2007).
Due to their high edge to volume ratio and high contact time of water in these areas, processes in
small streams are more sensitive to disturbance, including land use and climate change (Bishop et al.
Chapter Two 17
2008; Lowe and Likens 2005). Therefore, knowledge of the small streams that make up the complex
river systems on the Canterbury Plains could make a tangible difference to management of water
quality as a whole.
The theory of landscape ecology holds that heterogeneous spatial patterns matter, as they set
the context for organic processes such as fluxes of organisms, matter and energy (Fortin and Dale
2005; Wu and Hobbs 2007). Early models of stream ecosystems such as the River Continuum
Concept (RCC) were limited in their spatial consideration (Vannote et al. 1980). The RCC
considered physical, chemical and biological attributes of a stream to be changing along longitudinal
gradients from headwaters to lower reaches (Vannote et al. 1980). Gradually, however, landscape
ecology ideas have moved into stream ecology (Lowe et al. 2006; Thompson and Lake 2009). The
serial discontinuity concept made some headway towards a concept of rivers as interrupted continua
(resulting from human alteration) (Ward and Stanford 1995). Recently multidimensional
perspectives of rivers have been developed and a more continuous view, concerning vertical and
lateral linkage between river channels and the adjacent subsystems (hyporheic, parafluvial and
riparian) along the river continuum, have now been included in various models (Fisher et al. 1998;
Fausch et al. 2002; Junk and Wantzen, 2004).
In this study, I employed a landscape ecology approach to examine the spatial patterning of
water quality in stream networks, with the goal of understanding how land-use effects proliferate
through complex river systems. I examined the spatial variability in physico-chemical water quality
within five stream networks on the Canterbury Plains of New Zealand. I first investigated how
stream size and position with a network affected physico-chemical water quality. I hypothesised that
streams of different size would exhibit a different range of water quality conditions. I went on to test
the hypothesis that the overall spatial patterns in water quality differed between systems. Spatial
patterns were assessed by examining the level of spatial similarity that existed between sites within
Chapter Two 18
each system. Various distance measures and system level properties were used to identify factors
that may govern spatial patterns.
2.2 METHODS
2.2.1 Site Selection and System Mapping
Canterbury has over 27 000 km of running river channels, more than any other region in New
Zealand (Winterbourn 2008). Five predominantly spring-fed river systems were selected in the East
Coast Plains Ecoregion of the Canterbury plains, which extends from the foothills of the Southern
Alps to the East Coast, excluding Banks Peninsula (as recognised by Harding and Winterbourn
1995) (Figure 1). The rivers are situated on the peripheries of alluvial fans produced by the
Waimakariri and Selwyn Rivers (Winterbourn 2008). All five systems are considered lowland fed
rivers, a group that typically have spring sources and little hydrological variability. Subsurface
seepage can provide a substantial proportion of flow and heavy rainfall results in freshes, but these
are rarely major (Biggs 1985, Winterbourn 2008). Soil patterns on the plains can be deduced from
geomorphic history and are more complex than previously thought. Remnants of old alluvial fans
are sometimes preserved in association with small rivers (Webb 2008). Canterbury’s weather is
affected by oceanic and mountain effects with precipitation spread approximately evenly across the
plains, 600-800mm annually, with no strong seasonal variation (Sturman 2008). Riparian species
naturally occurring are flax, toetoe, grasses and sedges (Winterbourn 2008). Plains streams vary in
degree of naturalness, from stony-bottomed to drains that are channelised and often carry water from
elsewhere on the plains (Winterbourn 2008). Some streams are abundant in periphyton and rooted
macrophytes, which are home to a specific set of invertebrates and are periodically cleared by
councils and farmers (Winterbourn 2008). Since settlement in the 1850’s (Hawkins 1957), the
Chapter Two 19
Canterbury Plains have developed into what is now a patchwork of many types of agriculture,
lifestyle blocks and small towns. The systems I studied were selected with a preference towards
more complex dendricity, for the potential to reveal interesting spatial patterns. Three of the five
streams pass through small towns, and the Styx River through parts of Christchurch (Figure 1).
All waterways flowing into each of the five systems were mapped; this included all drains,
stock and irrigation raceways that joined the main stem. The flow path of water on the plains often
differs from that delineated on both topographical maps and the River Environments Classification
(REC), a GIS-derived database of network topology of New Zealand’s rivers and streams (Snelder
et al. 2005), due to the low elevation and complex alterations. Thus, I built a map of each river by
tracing the river lines using Google earth images (the most accurate and high quality images
available), in combination with field knowledge and ground truthing, using a GPS (Garmin GP560)
(Google Inc. 2009, GIS river lines created in ArcMap 9.2).
Sites were placed throughout each system as per the “snapshot” methodology, in order to
produce an instantaneous picture of all concentrations and fluxes in a watershed by sampling every
confluence and discharge point within a time period that is as short as possible and where flow is
stable (Salvia et al. 1999, Wayland et al. 2003, Walling and Webb 1975). Sites were placed where
feasible on all first order tributaries and point sources entering the system (sources with flow less
than 0.0005 m3s-1 were not sampled). They were situated at least 25 m above and below every point
of confluence, so that no two sites were less than 50 m apart. Sites were also placed along
uninterrupted main stem and tributary reaches, approximately 500 m apart (Figure 1). This method
of site selection resulted in systems having between 21 and 47 sites (Figure 1).
Chapter Two 20
Figure 1: The five stream networks studied were located on the Canterbury Plains of the South Island, New Zealand.
Shading in the upper left panel represents the topography of the area, the star is the location of central Christchurch and
the north arrow applies to all maps. The lower panels show the natural stream channels of five networks at and the dots
show the locations of sampling sites. (Note the different scales in each panel). River lines were constructed by tracing
from Google Earth images and ground truthing (Google Inc. 2009). The river lines and sampling sites are overlain on a
topographical map of the area, on which urban areas are shown by darker shading.
Chapter Two 21
2.2.2 Physico-chemical water quality sampling
Sites were sampled between March and April 2009, during base flow and were visited on a stream
by stream basis, sampling sites close together along a network on the same day so as to minimise
temporal variation in the data from each system. Base flow was defined as the period between
storms when the hydrograph is in the later stages of the recession limb usually 2 days post storm
flow peak (Pionke et al. 1999). Thus, sampling was not conducted if rain had occurred in the
previous 3 days in the catchment, Southern Alps or foothills, which may cause the Waimakariri,
Ashley and Selwyn rivers to rise, raising the water table feeding springs that contribute to several of
the systems. Each day I visited downstream sites first and collected water samples immediately on
arrival at a site to avoid contamination from sediments disturbed by entering the water. Sampling
was conducted up to 5 hours either side of midday in order to minimise diurnal variation in
parameters such as dissolved oxygen and temperature. Diurnal variation in water chemistry is
minimal with regards to all major forms of nitrogen and phosphorus except for ammonia (not
analysed here) (Finlay et al. 2011). In keeping with the aforementioned criteria, sites within each
system were sampled over a period of 3 to 10 days. To understand variation in water quality
throughout entire stream systems, one must sample in a spatially intensive way. In doing so, I accept
that my study could only be a “snapshot” in time, as a large-scale, spatially intensive study needs
many sites to be effective and my sites could only be sampled once.
At each site a 20-m reach was chosen to represent the general condition of the wider reach,
measured along the thalweg of the stream channel. I measured 8 physical and chemical properties of
the water at each site, using a YSI Sonde 6600, a multi-parameter water quality meter to
continuously log information at each site and taking water samples on ice for laboratory analysis.
Discharge and wetted width were measured at 3 points along the reach. Due to the large number of
sites sampled, a semi-qualitative approach was chosen to assess the physical habitat at each site
Chapter Two 22
(Harding et al. 2009). Physical in-stream properties assessed throughout the 20-m reach were algal
and macrophyte cover and a substrate index. An assessment of riparian conditions was also
undertaken at each site. Detailed field and laboratory methodology is provided in Table 1.
Chapter Two 23Table 1: Field and laboratory methodology for the eight water quality variables and other predictor variables measured at each site, as well as treatment in analyses.
Variable Units Field and/or laboratory methodology Reference Treatment Temperature ºC Time averaged spot measurement. The YSI Sonde 6600, a multi-parameter, water quality
measurement device was placed in the thalweg of a run at the top of each site, continuously logging during time spent at each site. The final 5-20 minutes of data from each site was averaged (visit times varied depending on sample run intensity).
I collected 80 mL of water in a syringe from the thalweg of the stream just below the surface and pushed through a filter (Whatman GFF 250 Millipore rating) into a 100mL opaque plastic bottle (pre-soaked in 5% hydrochloric acid overnight then rinsed three times with distilled water and a further three times with milli-q water). 80 mL of unfiltered water was collected in a similar way. Samples were kept on ice and then frozen. Filtered samples were thawed then analysed colourmetrically for DIN (assumed from combined nitrate-nitrite due to low nitrate concentrations). SRP using an automated, high throughput, water chemistry machine, the Easychem Plus (Systea, Italy). All chemicals used to make standards and reagents were of reagent grade and methods used by machine standard. The unfiltered water samples were thawed then processed for total phosphorus (TP) using standard colourmetric methods and testing for absorbance on a Trilogy Laboratory Fluorometer using a PO4 module.
American Public Health Association (APHA), 1995
Soluble reactive phosphorus (SRP) µg/L
Total phosphorus (TP) µg/L Wetzel et al. 2000
Riparian score Semi-quantitative protocol using P2d – Riparian procedure.
Harding et al. 2009
Predictor variables
Substrate index % A metric based on data gathered using P2c – In-stream habitat procedure. Algal cover % Visual estimate of % of the stream bed covered by algae. Macrophyte cover % Visual estimation of % of the stream bed covered by macrophytes. Shading % Visual estimate of % of the stream bed shaded when the sun is overhead.
Discharge m³/s
Discharge measured across one run, evenly flowing and free of obstructions. Offsets were placed wherever depth or discharge changed noticeably, with no fewer than five per transect. Water depth read on the downstream side of a ruler and water velocity is measured four-tenths of water depth up from the bed using a Marsh-McBirney Flow Mate. Discharge was calculated based on standard methods.
Gordon et al. 2004,
Harding et al. 2009
Wetted Width m Wetted width was measured at five locations along the 20m reach. Harding et al. 2009
Chapter Two 24
2.2.3 Land use, riparian conditions and other GIS derived variables
For each system, catchment area was defined using the River Environment Classification (REC)
(Snelder et al. 2005). Due to the low topography of the Plains, catchments generated from digital
elevations only approximate actual catchment areas, so manual changes were made to fit known
water movement through ground truthing. This is further complicated by stock water races and
irrigation canals, which result in considerable cross-catchment transfer of water. To estimate the
relative inputs of these artificial waterways between sites, an upstream distance measure which
included their length was calculated, as well as their proportional contribution to the overall
discharge from each system. The elevation change in each catchment was also derived from the
REC (Snelder et al. 2005), while stream order and the number of junctions were counted manually.
All parameters related to stream channel distance are based on the manually constructed stream
maps. Measures of network dendricity included drainage and confluence density (Benda et al.
2004). The contribution to each system by ground water was derived as the number of springs and
proportion of confined aquifer in the catchment area (Environment Canterbury GIS Layers).The area
of each catchment situated within a 100-m buffer of the stream channel was also calculated. For
each catchment and 100-m buffer zone metrics of proportional land use, forest presence, urban
development and road density were derived.
2.2.4 Distance Measures
Various matrices of the distance between each site and every other site in each network were
constructed. The Euclidean pair-wise distance was calculated between the sites in each system from
the NZMG (New Zealand Map Grid) coordinates of each site. This was calculated in R using that
spatial Statistics add on (spatstat, R Development Core Team 2007) to create a Euclidean distance
matrix, representing direct or overland distance between sites. A stream distance matrix was
Chapter Two 25
calculated using the mapped streams and the spatial analyst function to better represent the actual
path water takes across the landscape (ArcMap 9.2). To account for variable connectivity, a binary
connectivity matrix was constructed for each system, with sites that were connected directly by flow
attributed a one and all other sites attributed zero (Dray et al. 2006). For example sites on two
separate first order tributaries of the main stem are connected to all sites downstream on the main
stem and sites upstream on their individual branches, however not by flow to each other. Finally a
stream distance matrix, which was weighted by the inverse of the average flow encountered at sites
between any pair of sites, was constructed to better represent the ease of transfer of solutes and
particles between sites.
Chapter Two 26Table 2. List of all system wide metrics and method of derivation, calculated for each of the five systems as a whole.
System wide variables Methodology/calculation (all variables are based on the stream maps I constructed) Units Natural stream length The length of stream (including all tributaries) in the natural catchment upstream of the lowest site based on
traced and ground truthed stream maps. The decision was made to “cut off” a natural stream reach (beginning the “unnatural” portion), when upstream reaches took the form of highly channelised, free-flowing canals, of uniform depth, characteristic of water-race or irrigation networks (Google Inc. 2009, ArcMap 9.2).
m
Length of all waterways entering a system The length of stream including all natural and man-made waterways upstream of the site. Where branches split in two in the direction of flow, I only measured the branch connecting to the system. (Waimakariri District Council water-race data in geospatial form and ground truthing).
m
Proportion of network that is water-race fed The combined volume of water entering the system (instantaneous discharge) at points of “cut off” from the natural network, expressed as a proportion of the discharge of the most downstream site.
Total number of junctions Counted manually (excluding the water-race network). Number of springs in catchment stream Calculated using the Environment Canterbury online GIS database, springs found to date by Ecan field
workers, (Environment Canterbury GIS layers).
Catchment area Catchment areas were defined using the River Environments Classification (REC), then altered manually to meet known directions of water flow and extra channels not delineated by this model (Snelder et al. 2005).
km2
Area of 100-m buffer zone The area upstream of each site within 100m of the natural stream. This distance was chosen based on its use in buffer zone delineation (Baker et al. 2007) and as the smallest distance class to measure land use metrics to assess critical distances of impact (Houlahan and Findlay, 2004) (ArcMap 9.2).
km2
Length of road per km2 of catchment area The total length of all paved and metalled roads in a catchment area (New Zealand Landcover Database ver.2, Terralink 2004) expressed as a fraction of the area in km2.
Proportion catchment area that is urban The proportion of the catchment area that is built-up (New Zealand Landcover Database ver.2, Terralink 2004))
Proportion buffer zone that is urban As above but for the 100-m buffer zone. Proportion catchment that is in moderate to high intensity farming
The proportion catchment area that is in use as Dairy, Beef, Sheep or Sheep and Beef (Agribase, 2009)
Proportion buffer zone that is in moderate to high intensity farming
As above but for the 100-m near zone.
Proportion catchment that is in dairy farming The proportion catchment area that is in use as Dairy (Agribase, 2009) Proportion buffer zone that is in dairy farming As above but for the 100-m buffer zone. Proportion catchment with urban cover The proportion of the natural upstream catchment that is built-up (New Zealand Landcover Database ver.2,
Terralink 2004))
Proportion buffer zone with urban cover As above but for the 100-m buffer zone. Elevation change within catchment. Based on the reach of highest and lowest elevation in all mapped stream reaches that correspond to REC
reaches (Snelder et al. 2005). m
Proportion catchment with forest cover The proportion of the catchment covered in exotic or native forest (New Zealand Landcover Database ver.2, (Terralink 2004))
Proportion buffer zone with forest cover As above but for the 100m buffer zone. Drainage density Length of natural stream channel per km2 catchment area (Benda et al. 2004) Confluence density The number of confluences/junctions per km2 catchment area (Benda et al. 2004) Proportion of catchment area that on a confined aquifer
Estimated visually from maps of aquifer locations and known catchment areas (Environment Canterbury GIS layers).
Chapter Two 27
2.2.5 Analyses
2.2.5.1 Variance in physico-chemical variables associated with stream size
An initial overview of patterns was achieved by plotting box and whisker graphs of each physico-
chemical variable against stream order. The variability of the physico-chemical data was analysed
by plotting the PCA axis one and two scores from an ordination of all physico-chemical variables,
against stream wetted width (variables were transformed to meet normality assumptions), centered
and standardised before ordination as is usual for environmental data (Clarke and Corley 2006).
Quantile regression was performed to fit regression lines that bound the upper and lower five
percent of the data in the R package quantreg using the "br" method, which is a variant of the
Barrodale and Roberts (1974) simplex algorithm and is suitable for data sets of this size (Koenker
2009). These were tested for significance using the "boot" method to compute standard errors
(Koenker 2009).
2.2.5.2 Spatial patterns in water quality
To initially explore system wide patterns in water quality, each parameter was expressed visually
using graduated color symbols to represent the value of each site, at its actual location. To
investigate spatial patterns in physico-chemical variables I first tested how spatially structured the
water quality was within each system and compared across the five systems. Secondly the four
different distance metrics were used to tease apart which type of spatial patterns may be acting in
each system. Finally, I examined the cause of spatial patterns, comparing the level of spatial
structuring in each system to a range of system level properties.
Chapter Two 28
2.2.5.3 Spatial autocorrelation within systems
To examine the extent of spatial structuring within each system I constructed a series of spatial
eigenvectors, based on principal coordinates of neighbor matrices (PCNM) and related
methodologies. These methods are based on and comparable to the Moran’s I statistics, which are
the most commonly used statistics for autocorrelation analysis (Rangel et al. 2006). Eigenfunction
analysis produces a set of eigenvectors which each represent an orthogonal spatial structure (as they
are the product of a symmetric matrix) (Dray et al. 2006). I used eigenfunction-based spatial
filtering techniques to evaluate how well different distance measures accounted for variance in water
quality and in-stream conditions as a multivariate suite and individually (Rangel et al. 2006, Griffith
and Peres-Neto 2006, Blanchet et al. 2008b).
PCNMs code spatial information in a way that allows one to recover various structures over
the whole range of scales that the sampling design passes (Borcard and Legendre 2002). The starting
point of the PCNM approach is close neighborhood relationships among sites. First a truncated
distance matrix is constructed among sites, using a threshold value (defined as the minimum
distance that keeps all sites connected, based on a minimum spanning tree algorithm (Laliberté et al.
2008)), under which all distances are kept as measured and above which they are considered “large”
and an arbitrary value is replaced. In the second step, principal coordinates analysis is computed on
the modified distance matrix, necessary to represent the spatial information in a form compatible
with multiple regression and canonical redundancy analysis (RDA) (Borcard and Legendre 2002).
One, or several null and several negative eigenvalues are obtained, these cannot be used as they
correspond to complex numbers, however, the positive eigenvalues represent the Euclidean
components of the neighborhood relationships. Empirical results show that the positive eigenvalues
alone give a good representation of the spatial relationships (Borcard and Legendre 2002, Borcard et
al. 2004). I used both Euclidean distance (PCNM-E) and stream channel distance (PCNM-S), to
Chapter Two 29
construct two sets of spatial eigenvectors using the basic PCNM methods (Table 3), to test the
importance of overland, direct distances and in-stream distances on structuring water quality and in-
stream conditions, considering all sites as connected (Table 4 and see Table 3 for distance matrix
construction methods).
This basic framework has been developed to include options for directional and weighted
spatial representations with the potential to accurately represent processes in stream networks
(Blanchet et al. 2008a, Dray et al. 2006). Another set of spatial eigenvectors were created using the
Moran’s eigenvector map (MEM) method (Table 3). This involved the diagonalisation of a spatial
weighting matrix, constructed through the Hadamard product between two previously computed
resemblance matrices, a binary connectivity matrix and a weighting matrix (stream distance between
sites) (Dray et al. 2006). This allowed variable connectivity and ease of travel between sites to be
taken into account in the construction of this metric (MEM, Table 4). Finally I used average
discharge to weight the links between sites to represent the rate at which solutes and particles in the
water are transferred throughout the system. The weighted links table was then multiplied by the
matrix of stream distances (as in MEM with the connectance matrix) to create corresponding
weighted metrics (MEM-W Table 3). Eigenvectors (constructed by any method) with large
eigenvalues describe global structures whereas those with small eigenvalues describe local
structures (Borcard and Legendre 2002).
The spatial metric that best described variation in water quality and in-stream conditions was
chosen using the adjusted coefficient of multiple determination (R2a) to compare the variance
explained by each (Peres-Neto et al. 2006, Blanchet et al. 2008a). Each set of eigenvectors from
each of the eigenfunction analyses for each distance measure, was subjected to forward selection (α
< 0.1) to detect eigenvectors explaining the most variance in water quality (Blanchet et al. 2008a).
The multivariate set of physico-chemical water quality variables was then analysed as functions of
the set spatial eigenvectors by canonical redundancy analysis (RDA), a multivariate regression-
Chapter Two 30
based analysis using the spatial vectors as predictors of water quality (Dray et al. 2006, Peres-Neto
and Legendre 2010). The spatial metrics were then assessed as to their ability to describe spatial
structures within each system by comparing the R2a. The level of spatial structuring was also
compared between the five systems using the R2a associated with the best spatial descriptive metric
(Blanchet et al. 2008a).
These analyses were conducted in the R-language environment (R Development Core Team
2007) using the packages “vegan” (Oksanen et al. 2007) for RDA and variation partitioning,
“PCMN” (Dray et al. 2006) for the construction of PCNM variables and “packfor” (Dray 2005) for
the selection of explanatory variables in the RDA. In all tests of significance, 999 permutations were
used. Following Anderson and Legendre (1999) permutation of raw data were adequate for ANOVA
as there are no outlier values in the factors.
Table 3. Four different spatial metrics were tested on sites in each of the five systems. Eigenfunction-based spatial
filtering techniques were used, allowing flexibility in weighting and directionality of spatial representation. I used
PCNM with two distance metrics, Euclidean distance and stream distance. A third spatial metric based on Moran’s
eigenvector maps (MEM) used stream distance between flow-connected sites to account for the dendritic nature of site-
wise connectivity. A corresponding weighted metric was constructed based on the average velocity encountered between
sites.
Distance metric Weight Code DescriptionEuclidean distance PCNM-E PCNM on Euclidean distances between
sites
Stream distance PCNM-S PCNM on stream distances between sites
Stream distance between flow connected sites only
MEM MEM all sites directly connected by flow in both directions
Average discharge
MEM-W MEM – weighted by average discharge encountered between sites.
2.2.5.4 Potential drivers of system level spatial autocorrelation
The R2a associated with the best spatial descriptive metric was considered to represent the level of
spatial autocorrelation within each system. To investigate the causes of varying levels of spatial
Chapter Two 31
structure in stream water quality, the best R2a each network was compared to a series of system level
parameters (Table 2) using simple linear regression in the R-language environment (R Development
Core Team 2007).
2.3 RESULTS
2.3.1 Variance in physico-chemical variables associated with stream size
No strong trends were evident in median water quality value between the four stream orders
examined, when compared across all sites and systems. Due to the non-independence and lack of
normality of the data p values were not tested statistically. It is clear, however, that there was more
variation about the median in the first and second order sites than the third and fourth, across all
variables (Figure 2a-f).
Chapter Two 32S
RP
(µg
L-1)
0
25
50
75
100
TP (µ
g L-1
)
0
200
400
600
Stream Order
1 2 3 4
SIN
(mg
L-1)
0
4
8
12
Stream Order
1 2 3 4
Tem
pera
ture
ºC
10
12
14
16
18
Dis
solv
ed O
xyge
n (m
g L-1
)
2
4
6
8
10
12
14
0
100
200
300
400
Con
duct
ivity
(µS
)
(a)
(b)
(c)
(d)
(e)
(f)
Figure 2. Variation in soluble reactive phosphorus (a), total phosphorus (b), nitrate (c), conductivity (e), dissolved
oxygen (e) and temperature (f) across all sites and systems by stream order. The shaded boxes show the spread of data to
the 25th and 75th percentiles around the median value, lines extend out to the 5th and 95th percentiles and dots are outliers.
Quantile regression was used to evaluate the variance structure across the range of the data
set. Variability in physico-chemical PCA axis one and two scores declined with increasing stream
width (Figure 3). Axis one described the variability in dissolved oxygen, pH, temperature and
Chapter Two 33
conductivity. The slope of the 10th quantile regression line is significantly different from zero,
indicating a positive floor, limit relationship (p = 0.005, Figure 3a). The variation in the upper level
of axis two scores, which represented concentration of dissolved oxygen, SRP, TP and turbidity,
significantly reduced with increasing stream size (p =. 0.01755, Figure 3b). Water quality in small
streams was highly variable when compared with the homogeneous nature of larger streams.
Wetted width of stream (m)
0 5 10 15
Axis
one
sco
res
from
PC
A of
phy
sico
chem
ical
var
iabl
es
Dis
solv
ed O
xyge
n
pH
, tem
pera
ture
, con
duct
ivty
, NO
x
Wetted width of stream (m)
0 5 10 15
A
xis
two
scor
es fr
om P
CA
of p
hysi
coch
emic
al v
aria
bles
D
isso
lved
Oxy
gen
TP
, SR
P a
nd T
urbi
dity
(a)
(b)
Figure 3. Scatter plot and quantile regression fits of the principal components analysis axis one and two scores, from the
Standardised PCA of the physico-chemical variables against stream wetted width. Variables significantly correlated with
each axis are displayed. Superimposed are the 10th (a) and 90th (b) quantile regression lines.
Chapter Two 34
2.3.2 Spatial patterns in system-wide water quality
The panels below show the high level of variability in water quality that exists within a
single system, and between two systems (Figure 4). Taking Birdlings Brook and the Cam River as
examples, one can see that each system demonstrates patterns that are unique, with respect to each
physico-chemical variable, and that these patterns are not always consistent between these two
systems (Figure 4). Dissolved oxygen shows lower levels in the tributaries than main stems of
Birdlings Brook, whereas branch specific dissolved oxygen levels occur in the Cam River (Figure
4a). Total phosphorus decreased down the main stems of Birdlings Brook, yet, comparably, is low
throughout the Cam River. A tributary high in total phosphorus enters the Coldstream branch of the
Cam River, yet this branch remains low in total phosphorus below this confluence (Figure 4b). DIN
had a patchy distribution in Birdlings Brook, while again in the Cam River the distribution was
branch specific (Figure 4c)
Chapter Two 35
Figure 4. Graduated symbols represent spatial variation in dissolved oxygen (a and b), total phosphourous (c and d) and
dissolved nitrate (e and f) at each site in Birdlings Brook (left panels) and the Cam River (right panels). The same key is
relevant across both systems for each variable. Sites are overlain on the mapped streamlines and topographic map
images. The catagories of each variable were defined using the “Jenks” natural break method (ArcMap 9.2).
Chapter Two 36
2.3.3 Comparing metrics and levels of spatial autocorrelation
The spatial variables created by MEM method described the most spatial variation in three of the
five systems, while those created by the PCNM method using stream distance performed the best in
the remaining two systems (Table 5, Figure 5). Either MEM or MEM-W performed relatively well
in all systems (Table 5, Figure 5). However the order of performance of the spatial metrics differed
between the systems. Euclidean distance generally explained relatively little spatial variation,
however in Birdlings Brook and the L2 it did not perform quite as poorly as in the other systems (by
comparison within each) (Table 5, Figure 5). Birdlings Brook had the highest level of spatial
structuring of all five systems (the largest R2a values across all metrics), followed by the Cam River.
This means that sites that are close together have similar levels of water quality, compared with sites
that are far apart. The Styx River had the least amount of spatial structuring in water quality, with no
significant spatial vectors being chosen at all from the PCNM-E method (Table 5). Hunters Stream
differs from the other systems in that the metrics are not evenly spread in their ability to describe the
spatial pattern. Despite having the least spatial pattern in general, described by three of the metrics,
Hunters Stream displays a moderate amount of spatial structure compared to the other systems,
when described by the MEM weighted metric (Table 5, Figure 5).
In general the significant spatial PCNM vectors are fewer and describe larger spatial
structures than the MEM vectors, which cover a broader spatial range (Table 5). The L2 had spatial
structuring at the finest scale, as well as at a large scale, while Hunters Stream only at the large scale
and the Styx only at an intermediate scale (Table 5). Birdlings Brook and the Cam River had
significant spatial vectors that described a wide spread of spatial scales (Table 5).
Chapter Two 37Table 5. Differing amounts of variance in water quality composition were explained by the four spatial metrics in the
five stream networks. This was determined using redundancy analysis (RDA) with forward selection (α <0.1 and 9999
permutations). The ‘# sig. vectors’ indicates the number of vectors that were significant during forward selection, while
‘variance’ is how much variation in community composition this set of variables explained. The most significant
distance metric for each stream is highlighted in bold with probability (P) values indicated. R2a is the correlation
coefficient, adjusted for the number of variables in each model so that they can be compared robustly. Refer to Table 2
for distance measurement acronyms.
# sig.
vectors Variance P R2a
Ranges of sig. spatial vectors
(eigenvalues on Log scale)
L2
PCNM-E 2 0.20 0.015 0.13
PCNM-S 3 0.34 0.005 0.25
MEM 7 0.42 0.015 0.20
MEM-W 3 0.17 0.11 0.06
Birdlings Brook
PCNM-E 5 0.57 0.015 0.43
PCNM-S 2 0.41 0.005 0.34
MEM 9 0.82 0.005 0.67
MEM-W 8 0.88 0.005 0.81
Hunters Stream
PCNM-E 1 0.08 0.16 0.03
PCNM-S 2 0.17 0.053 0.09
MEM 2 0.11 0.17 0.04
MEM-W 10 0.65 0.005 0.39
Cam River
PCNM-E 5 0.30 0.005 0.22
PCNM-S 7 0.36 0.005 0.26
MEM 21 0.71 0.005 0.46
MEM-W 26 0.76 0.005 0.47
Styx River
PCNM-E 0 NA NA NA
PCNM-S 3 0.31 0.026 0.19
MEM 2 0.26 0.01 0.18
MEM-W 2 0.25 0.015 0.17
Chapter Two 38
Cam Birdlings Hunters L2 Styx
R2 a
0.0
0.2
0.4
0.6
0.8
1.0
PCNM Euclidian DistancePCNM Stream distacneMEM MEM Velocity Weighted
Figure 5. Differing amounts of variance in water quality were explained by the four spatial metrics in the five stream
networks. This was determined using redundancy analysis (RDA) with forward selection (α <0.1 and 9999
permutations). R2a is the correlation coefficient, adjusted for the number of variables in each model so that they can be
compared robustly. Refer to Table 2 for distance measurement acronyms and text for detailed methods.
2.3.4 Testing for spatial autocorreletaion in the systems
A broad range of system level properties were tested against the level of spatial autocorrelation
present in each system, described by the R2a value associated with spatial metric that described the
most variance in water quality. The amount of variance in water quality explained by the best spatial
model in each system significantly decreased as catchment area increased (R2 = 0.81, p = 0.037,
Figure 6d). The proportion of forest cover in the 100m-buffer zone and proportion of the catchment
area that was confined aquifer also had negative relationship with of variance in water quality
explained spatially, however these were weaker relationships than that of catchment area, significant
at α=0.1 (R2 = 0.56, p = 0.087 and R2 = 0.3749, p = 0.097 respectively, Figure 6a and b). The
number of springs in each catchment area, and both measures of dendricity, (drainage and
Chapter Two 39
confluence density) had no significant influence on the level of spatial autocorrelation in the five
systems I studied (Figure 6c, e and f).
Proportion of 100m-buffer zone with forest cover0.00 0.01 0.02 0.03 0.04
0.0
0.2
0.4
0.6
0.8
1.0
Proportion of catchment area in confined aquifer0.0 0.2 0.4 0.6 0.8 1.0 1.2
R2 a
0.0
0.2
0.4
0.6
0.8
1.0
Number of springs arising within the catchment area0 50 100 150
0.0
0.2
0.4
0.6
0.8
1.0
R2 a
R2 a
Catchment areas (km2)
15 20 25 300.0
0.2
0.4
0.6
0.8
1.0
Drainage density1.4 1.6 1.8 2.0 2.2
0.0
0.2
0.4
0.6
0.8
1.0
Confluence density0.0 0.5 1.0 1.5 2.0
0.0
0.2
0.4
0.6
0.8
1.0
(a)
(b)
(c)
(d)
(e)
(f)
Figure 6. The amounts of variance in water quality explained by the best spatial metrics (using the associated R2
a) in
each of the five stream networks compared with system level properties, proportion of the 100-m buffer zone with
forest cover (a), proportion of catchment area in confined aquifer (b), number of springs within each catchment (c),
catchment area (d) and confluence and drainage densities (e and f) (see table 2 for derivation of these). Regression
lines displayed where the general liner model indicted a significant relationship (the solid line is significant at α=0.05
and dashed lines at α=0.1)
Chapter Two 40
2.4 DISCUSSION
Water quality is an important metric of stream health; however, variability of surrounding
land-use and innate features of the stream network alter the pattern of water quality in a system. I
found that headwater streams were most affected by surrounding conditions and had the greatest
variability in water quality. Network characteristics in the five stream systems led to different levels
in spatial structure between streams. Specifically, spatial variation in nutrient inputs (via
groundwater inputs and runoff) and in-stream nutrient processing may alter the degree of spatial
autocorrelation within stream networks. The degree of spatial structure will affect how headwater
effects on water quality are proliferated through the stream network.
2.4.1 Pervasive variance in the water quality of small upstream sites
Using several metrics of stream size, I found that small stream sites had more variable water quality
than larger stream sites. Thus, by extension, sites on tributaries, further up the network were more
variable than main stem sites, nearer the outflow of the network. A similar trend was found for five
out of the eight water quality variables in low elevation, pastoral streams, from throughout New
Zealand (Larned et al. 2004) and in long term monitoring of the Seine River in France, where the
range in conductivity values decreased with increasing stream order (Meybeck et al. 1999). The
change in variance structure with stream size may help explain why limnological data rarely
conform to parametric assumptions (Matthews et al. 1991). For this reason, systematic trends and
causal relationships with stream size have often been difficult to detect, even when present in water
quality data (McKee et al. 2001, Larned et al. 2004). Heterogeneous errors seem to be pervasive in
water quality data, resulting in the need for transformation and high levels of unexplained variation
(Jones et al. 2001, McKee et al. 2001, Larned et al. 2004). The identification of gradients along
Chapter Two 41
streams, by order, has been successful for biotic parameters, but often not for water quality (Beecher
et al. 1988, Naiman et al. 1997). The pervasiveness of this variance is demonstrated in heterogeneity
in water quality that occurs within single catchments, in pristine condition (Clow et al. 1996, Eyre
and Pepperell 1999, Finlay et al. 2011).
Water quality in headwater streams is heavily influenced by the nature of the surrounding
landscape, the characteristics of local and regional groundwater inputs, degree of riparian buffering,
and in-stream processes. Due to their small size, headwater streams are particularly sensitive to
variations in these controlling factors. The nature of stream networks in the Canterbury lowlands
(i.e. dendritic network morphology) is such that headwater streams are widely distributed
throughout the landscape and are exposed to a wide range of land-use, hydrologic and biological
conditions. This higher sensitivity and wider exposure lead to the considerable variation in water
quality that I observed in headwater streams.
Small streams are more strongly influenced by local land-use conditions than larger ones
(Buck et al. 2004), this is likely to enhance their variability (Bishop et al. 2008). This can be
explained by the higher edge to volume ratio compared with large streams (Bishop et al. 2008). This
is one of the principles behind sampling water quality at base-flow, when the mean transit time of
water (thus, contact with the landscape) is highest at low flow; so it is thought that water quality will
be tightly coupled with land use (Biggs et al. 1990, Lyon et al. 2008). In small streams, the smaller
volume of water is less able to dilute the diffuse influences of surrounding land and any point source
influences, compared with larger discharge sites. Whereas, sites further down a system, with a
higher flow volume, are more able to buffer or dilute the impacts of incoming polluted tributaries
(for example, the tributary high in total phosphorus (discussed in the results) has no impact at
downstream mainstem sites).
A further explanation for the small stream variability I found on the Canterbury Plains is the
general heterogeneity of landscape conditions through which each network passes. Despite the flat
Chapter Two 42
appearance of the Canterbury Plains, with increasing urban sprawl, the dividing up of large farms
into smaller mixed-use land and lifestyle blocks, as well as pockets of vegetation restoration, the
Plains are far from homogenous. Soil pattern on the plains can be deduced from geomorphic history
and are more complex than previously thought. Remnants of old alluvial fans are sometimes
preserved in association with small rivers (Webb 2008) and many soil types exist within each of the
five catchments studied (Environment Canterbury GIS layer). This could lead water to come in
contact with sediment and soil of variable characteristics and redox conditions can alter
biogeochemical pathways and the patterns of stream water quality (Dahm et al. 1998). Additionally,
different types of land-uses can have unique influences on individual water quality parameters
(Jones et al. 2001, Jones et al. 2004) and extremes of each variable are more likely to exist in first-
order streams. In small streams, the contact time between nutrients and the water column is also
high, due to the low flow rate. Thus, as well as being perceptive of the local land use, small streams
can be particularly effective in retaining nutrients due to shorter uptake length for nutrient spiralling
compared with large streams (Peterson et al. 2001, Niyogi et al. 2004). Thus, their variability in
biological conditions may lead to spatial and temporal variability in water quality of lower stream
orders.
Large scale variation is also likely to exist in the interaction of surface and ground waters on
the Plains. Confined and unconfined aquifer matrices exist in a spatially variable, alternating form
that has been described as “beads on a string” (Cruezé des Châtelliers et al. 1994). Many
microclimates exist on the Canterbury Plains that could potentially add to and exacerbate the effects
of groundwater variability (Sturman 2008). Given that a significant amount of variability has been
found in the water chemistry of a single catchment, with no bedrock variation and or human impact
(Clow et al. 1996), it is reasonable to expect much variability on the Canterbury Plains.
Adding to this effect, riparian management is inconsistently practised across the Canterbury
Plains (Greenwood, McIntosh and Harding, unpublished data). Small streams are often unfenced to
Chapter Two 43
allow for stock access to drinking water and have had any remaining riparian buffer removed to
increase productive land. Stream banks often collapse adding sediment and enhancing transport of
pollutants directly into the waterways (Tabacchi et al., 1998, Parkyn et al., 2003, Baker et al. 2007).
Several studies have found that surface runoff only enters a river system through a relatively small
proportion of the total riparian length (McGlynn and Seibert, 2003) and have observed decreases in
surface runoff with increasing stream order. This, along with their unbalanced contribution to
system length, has lead several authors to conclude that riparian zones along headwater streams may
be disproportionately important for nutrient retention (Dosskey et al. 2005).
2.4.2 Identifying system level patterns using spatial autocorrelation – untangling the
variance
A tendency towards positive spatial autocorrelation at small scales makes intuitive sense. Sites that
are close together on the stream are similar. Sites begin to show no spatial autocorrelation
(randomness) or negative spatial autocorrelation (homogeneity) once they are a certain distance
apart (Urban, 2003). Measures of spatial autocorrelation are a useful way of summarising variable
spatial patterns. Some level of spatial autocorrelation is present in most stream water quality data
due to the directional nature of flowing water (Lyon et al. 2008, Tu and Xia 2008). The presence of
a connection by flow and the distance along the stream channel between sites potentially has a
greater influence on spatial patterns in water quality than the distance directly, or “as the crow flies”,
between sites (Fagan 2002). This was the case in all five systems I studied, where distance metrics
based on stream channel distance consistently explained more variance in the spatial pattern than the
direct (Euclidean) distance. Therefore, I conclude, as was found in several recent studies, that stream
channel distances are more suitable for use in water quality studies than Euclidean distances.
Euclidean distance fails to accurately represent the spatial configuration, connectivity and relative
position of sites in a stream network (Peterson et al. 2007, Lyon et al. 2008).
Chapter Two 44
Increasing heterogeneity of water quality with increasing separation between sites along a
stream can arise from changes in the local land use and point source inputs, such as tributaries (but
also unidentified water entering the system - piped outfalls or ground water upwelling). Changes in
riparian and in-stream conditions (macrophyte or algal cover) may in turn lead to differing levels of
nutrient spiralling. Any disruption to the balance between these processes, leading to net retention or
regeneration of nutrients, may disrupt the level of pattern spatial similarity within a stream system.
Thus, the extent of spatial similarity within a system can be used to make inferences regarding the
occurrence of these processes.
One concept worth applying is that buffer zones can be either leaky or retentive, changing
the connectivity between the surrounding landscape and the stream (Baker et al. 2007). Vegetated
buffer zones can act as a sink for nutrients (Billen and Garnier 2000), insulating the stream channel
from the impacts of surrounding land use. Water quality would be expected to be more spatially
similar under such “closed” conditions, relying on in-stream processes to regulate water quality.
Reaches without intact riparian buffers are more “open” to impacts from the surrounding land,
relying more on terrestrial connectivity than aquatic biology. Such factors may alter water chemistry
at similar sites differently than would be expected for their degree of physical separation and chance
alone.
The high level of spatial structuring of the water quality within Birdlings Brook and the Cam
River could be due to the fact that these streams are well insulated from the surrounding landscape
and/or demonstrate balanced in-stream nutrient processes. The widely spread eigenvalues of the
spatial vectors, which best modelled the water quality variation in these systems, indicated that
processes may be balanced at multiple scales. However, my results indicated a negative relationship
between the amount of spatial autocorrelation and the proportion of the buffer zone that is forested.
At the scale of the Canterbury Plains, only coarse scale forest cover data is available, this may not
Chapter Two 45
be an accurate indicator of the actual riparian buffer conditions in these systems, particularly for
smaller streams.
The L2 and Styx Rivers have comparatively less spatial similarity in water quality than the
Cam River and Birdlings Brook. The L2 and Styx River systems flow through damp wetland areas
and, as they are situated on the edge of confined and unconfined aquifers, upwelling water produces
many springs that contribute to their flow. The spatial patterns may be less evident in these systems,
as they are interrupted by an influx of groundwater of different quality from the surface stream
water. Streams on the Canterbury Plains are likely to vary in their degree of openness with regard to
connectivity with groundwater. Across the five systems that I studied, the proportion of each
catchment that was on a confined aquifer explained the observed patterns, but the number of springs
did not.
Hunters Stream is unusual in that it exhibits very little significant spatial correlation, other
than when I examined it using the discharge weighted measure. The channels of this once-natural
stream are now fed almost entirely by water from stock water races. However, discharge decreases
longitudinally, which provides an explanation for the discharge weighted spatial metric being the
only metric that explained any significant level of the variation in this steam. The low flow levels in
Hunters Stream may mean that spatial patterns are only evident in those parts of the stream where
flow is higher. In these parts, the flow can transport nutrients or disturbance effects before the water
is lost from the channel. As the earlier results of this chapter indicate, water conditions are likely to
be highly variable in small streams, thus a variable pattern throughout the system may also result
from its nature as a small stream, based on the potential mechanisms discussed earlier.
One stream network characteristic that should have regulated spatial structure was branching
pattern. Specifically, I expected higher levels of dendricity would lead to decreases in the spatial
similarity of water quality. This decrease would potentially be caused by the confluence of many
tributaries, of differing water quality, disrupting longitudinal similarities. This effect would be
Chapter Two 46
exacerbated by the nature of dendritic systems, with their capillary of small streams, branching
throughout a landscape of potentially great variability (Bishop et al. 2008). Yet, in the five systems I
studied, there was no relationship between spatial autocorrelation in water quality and dendricity.
The exact cause of spatial structuring in water quality in each system considered in this study could
not be absolutely defined, as my analyses did not distinguish between the spatial dependence of
water quality (i.e. spatially structured environment) and the actual spatial processes driving
community structure (e.g. the downstream flow of solutes). However, my results confirm the
usefulness of the ‘space as a surrogate’ approach (McIntire and Fajardo 2009) for identifying areas
where landscape interaction or changes in in-stream conditions potentially occur. The interaction of
spatial processes with catchment land-use, riparian and in-stream processes, in influencing water
quality, was addressed using a focus on the Cam River in Chapters Two and Three.
2.4.3 The role of small streams in spatially correlated systems
In stream systems where water quality is spatially correlated, any disturbances at the headwaters
would proliferate through the system to a greater degree than in a system with less spatially
correlated conditions. My results indicated that the water quality of headwater streams can have a
disproportionate influence over water quality throughout an entire network. Care should be taken
during the intensification of land bordering small streams, as the impact of a disturbance may reach
far beyond the area of land under development. Whereas, the protection of headwater streams may
provide a cost-effective focus for management of catchment water quality, rather than protecting
entire systems. Management of small streams is particularly important, not only due to their spatial
control, but also due to their highly variable water quality within the lowland areas of the
Canterbury Plains.
Chapter Three 47
3 CHAPTER THREE - CONSISTENT SPATIO-TEMPORAL
WATER QUALITY VARIATION IN A COMPLEX RIVER
SYSTEM
The Northbrook: a branch of the Cam River, which flows through dairy farms to the East of
Rangiora.
Chapter Three 48
3.1 INTRODUCTION
The physical and chemical composition of river water strongly influences its suitability for both
aquatic life and use by humans (Davies-Colley and Willcock 2004). Early interest in water quality
arose from a geological perspective, in relation to catchment geolithology (Walling and Webb
1975). However, with increasing degradation of freshwaters by agriculture and urbanisation, both
worldwide and in New Zealand, it is important to understand variations in fresh water quality in
terms of multiple landscape variables (Jones et al. 2001, Allan 2004). Water quality is a valuable
indicator of the degree to which a river system has been impacted by changes in the environment
(Hem 1985, Jones et al. 2001). Yet several aspects of limnological data sets make them notoriously
difficult to deal with in analyses (Walling and Webb 1975, Matthews et al. 1991).
3.1.1 Difficulties in the analysis of water quality data
Limnological data are generally non-linear, rarely conform to parametric assumptions and are often
measured using incommensurable units such as length, concentration, and frequency (Matthews et
al. 1991). In addition, most water quality research generates incomplete data sets through sample
loss and sampling design (Matthews et al. 1991). Gradients and causal relationships may be difficult
to detect due to the often high variability in water quality data (McKee et al. 2001, Larned et al.
2004). Heterogeneity in water quality even occurs within single catchments in pristine condition
(Clow et al. 1996, Finlay et al. 2011). Stream water quality also varies temporally, largely through
changes in flow rate (Salvia et al. 1999, Pionke et al. 1999, Wayland et al. 2003).
Most, if not all, environmental data are also spatially correlated (Legendre and Fortin 1989).
However, this is a particularly significant problem in rivers, where the directional nature of flowing
water means that data collected are often spatially correlated and non-independent (Lyon et al. 2008,
Tu and Xia 2008). Two main problems have been identified. Firstly, spatial autocorrelation; where
Chapter Three 49
one site may have more similar water quality to a site nearby, than to another site far away
(Legendre 1993, Fortin and Dale 2005). This is because sites near each other may be affected by the
same human influences, climate, geology and may even be connected by flow. Secondly, spatial
non-stationarity; where the relationship between the independent and dependent variables are not
constant over space and time (Fotheringham et al. 2002). This often occurs in water quality data as
local conditions are likely to vary (Tu and Xia 2008). Relationships that exist in part of the study
area may be obscured by an apparent overall trend or lack thereof. These situations result in spatial
autocorrelation and heteroscedasticity in the residuals of analyses, which reduce their power to
detect causes of water quality variation (Legendre 2002).
Under positive spatial autocorrelation, the value of a parameter, at a point, can be predicted
in part from values at surrounding points. As values are not stochastically independent, in reality
each point brings less than 1 degree of freedom to the analysis, reducing its power (Legendre 1993,
Fortin and Dale 2005, Rangel et al. 2006). Thus, the assumptions of normality and homoscedasicity
are often ignored in water quality studies and many authors rely on the robustness of statistical tests
to identify significant trends, despite violation of fundamental assumptions (Matthews et al. 1991,
Tu and Xia 2008, Rangel et al. 2010).
Within stream networks, which are inherently complex, it is particularly difficult to detect
water quality relationships due to stochastic variability, natural and human induced spatial
heterogeneity and logistical challenges of working across multiple sites (Prowse 1984, Finlay et al.
2011). The high level of correlation between land use variables within a catchment is problematic
for their use in regression analysis (Jones et al. 2001). Water quality is usually monitored on a
regular basis at a small number of sites in a watershed, usually at the outlet of a larger basin, thus all
non-point and point sources are integrated and mixing of waters may obscure local extremes
(Walling and Web 1975, Salvia et al. 1999). Yet effective management involves identification of the
key sources and processes. Untangling the effects of geology, topography and land-use on spatial
Chapter Three 50
variation in water quality is also difficult, as these variables are often correlated, so one may obscure
the impact of another as in community ecology (Walling and Web 1975, Prowse 1984, Laliberté et
al. 2008). Temporal variations in water quality may also contain temporal autocorrelation, which
can further convolute studies which are conducted across space, but often through logistic
constraints, not conducted approximately simultaneously (Fortin and Dale 2005).
3.1.2 Investigating variability using a spatial-temporal snapshot
Stream ecologists have long studied biogeochemistry under a whole catchment perspective
(Webster et al. 1979, Deitrich et al. 1982). Recent work on the dendritic and hierarchical nature of
streams has highlighted the need to pay more attention to spatial structure and connectivity (France
and Duffy 2006, Ganio et al. 2005, Brown and Swan 2010). Moreover, many insights can be gained
by spatial studies of stream physicochemistry, especially if they incorporate temporal components.
Identification of the flow-on effects of changes across a wide range of scales, and predicting their
consequences, is an important topic for ecological research in general (Ludwig 2007). Not only did
early work on catchment level water quality variation fail to take these issues into account, elements
of spatial autocorrelation were overlooked in modelling and analysis (Brown and Swan 2010).
Although researchers of physico-chemical processes pioneered the inclusion of spatial perspective
(Hubbard Brook work and snapshot studies in the 1990’s), they have not benefited from the more
recent development of statistical methods capable of handling the problems associated with spatial
issues.
In all ecological fields, the interest in variability associated with landscape and space has
developed faster than the appropriate statistical methods (Rangel et al. 2006). A literature review
found that 80% of studies which analysed spatial data did not use space explicitly in their analysis
(Dormann 2007). Many researchers continue to use traditional statistical methods, despite
knowledge of the spatial nature of their data and new methodology developments (Herlihy et al.
Chapter Three 51
1998, Billen and Garnier, 2000, Jones et al. 2004, Finlay et al. 2011 and many more). Traditional
methods are not well equipped to deal with spatial autocorrelation, so water quality scientists risk
catchment have become dominated by dairy farming and lifestyle blocks (Biggs 1985, Agribase
2009).
All waterways flowing into the Cam River system were mapped; this included all drains and
water-races or irrigation canals that entered the system. The actual flow path of the Cam River
differs markedly from published delineations in several areas. Firstly the southernmost branch, as
described by topographical maps and the River Environment Classification (REC, Snelder et al. 2005)
does not flow into the Cam River, instead joins the Cust Main Drain. Secondly the water-race
network intersects the system near the source of the Southbrook and Northbrook. Ground truthing
revealed that this water-race network crossed the plains from the Waimakariri. An accurate GIS
layer of all water flowing into the Cam River was built by tracing the river lines using Google Earth
(Google Inc. 2009) images (the most accurate and high quality images available), in combination
with field knowledge and ground truthing, using a GPS (Garmin GP560) (Figure 1). The complex
dendricity and variety of land use within the Cam River catchment has the potential to reveal
interesting spatial patterns. In particular, the long Northbrook branch provided an opportunity to
study longitudinal effects on water quality.
3.2.2 Site selection
Base flows are most likely to be encountered in the late summer in New Zealand (Biggs et al. 1990).
I sampled water quality in the Cam River synoptically, on a monthly basis, from January to March,
in the austral summer of 2010. Sites were spread throughout the Cam River system as per the
“snapshot” methodology to produce an instantaneous picture of all concentrations and fluxes in a
watershed by sampling every confluence and discharge point within a short time period and where
flow was stable (Walling and Webb 1975, Grayson et al. 1997, Salvia et al. 1999). Sites were placed
where feasible on all first order tributaries and point sources entering the system (sources with flow
less than 0.0005 m3s-1 were not sampled). They were situated at least 25 m above and below every
Chapter Three 55
point of confluence, so that no two sites were less than 50 m apart. Sites were also placed along
uninterrupted main stem and tributary reaches, approximately 500 m apart (Figure 1). This method
of site selection resulted in 81 sites throughout the network, of which a subset were sampled in the
January and February sampling events, due to time constraints (Figure 1).
Figure 1: The location of the Cam River on the Canterbury Plains (a) and the location of the Cam River sampling sites
(b). The blue lines show the natural channels of the Cam River system (constructed by tracing from Google Earth
images and ground truthing). The stream drains towards the bottom right. The river lines and sampling sites are overlain
on a topographical map of the area. The nearby town Rangiora and three major branches of the river are identified, these
branches confluence in quick succession to the south of the town, forming the main Cam River
3.2.3 Snapshot Sampling
Samples were collected in three snapshot sampling events (Table 1), each spread over one to four
days within a period of base flow, defined as the period between storms when the hydrograph was in
the later stages of the recession limb, usually two days post storm flow peak (Pionke et al. 1999).
Thus, sampling was not conducted if rain had occurred in the previous week in the catchment as
well as in the Southern Alps or foothills which may cause the Waimakariri and Ashley rivers to rise,
Chapter Three 56
raising the water table feeding the springs in the Cam River. Measured stage height variation at three
sites indicates flow in Cam River was stable across each sampling period. Sampling was conducted
up to 5 hours either side of midday to minimise diurnal variation in parameters such as dissolved
oxygen and temperature. Diurnal variation is minimal for all major forms of nitrogen and
phosphorus except for ammonia (not analysed here) (Finlay et al. 2011).
Sampling was more intensive than the survey conducted in the previous summer (Chapter
Two) with 81 sites visited during the most intensive event. I measured eight physical and chemical
properties of the water, as well as discharge, at each visit to a site. Other in-stream and site
properties, which were considered more permanent, such as algal and macrophyte cover, were
assessed on only one occasion. All variables, as well as field and laboratory methodology are
described in Table 2. Each day I visited downstream sites first and collected water samples
immediately on arrival at a site to avoid contamination from sediments disturbed by entering the
water.
Chapter Three 57
Table 1: Summary of the information relating to the three sample events in 2010 and the one-off sample event in 2009. Data are averaged across the days of
sampling (NIWA, online climate data base, Rangiora station)
Average across sampling run days (air temperature analysed for sampling hours only)
No. sites
Sampling period
Discharge at outflow
Day since rain >2 mm
Total rain
Soil moisture
Mean air temp
Max air temp
Min air temp
Sunshine amount
Evaporation (Penman ET)
Units Day m³/s Day Mm % ºC ºC ºC Hours mm January (2010) 63 2 1.44 5 0 14.5 17.8 21.2 15.6 5.4 4 February (2010) 71 1 1.59 6 0 13.1 20.5 26.8 10.3 7.7 3.4 March (2010) 81 4 1.18 5 0.1 13.3 16.8 21.9 4.7 2 2.1 March (2009) 47 6 1.21 3 0.2 15.3 13.5 26.9 7.1 6.9 1.6
Chapter Three 58
Table 2: Field and laboratory methodology for the eight water quality variables and other predictor variables measured at each site, as well as treatment in analyses.
Variable Units Field and/or laboratory methodology Reference Treatment Temperature ºC Time averaged spot measurement. The YSI Sonde 6600, a multi-parameter, water quality
measurement device was placed in the thalweg of a run at the top of each site, continuously logging during time spent at each site. The final 5-20 minutes of data from each site was averaged (visit times varied depending on sample run intensity).
I collected 80 mL of water in a syringe from the thalweg of the stream just below the surface and pushed through a filter (Whatman GFF 250 Millipore rating) into a 100mL opaque plastic bottle (pre-soaked in 5% hydrochloric acid overnight then rinsed three times with distilled water and a further three times with milli-q water). 80 mL of unfiltered water was collected in a similar way. Samples were kept on ice and then frozen. Filtered samples were thawed then analysed colourmetrically for DIN (assumed from combined nitrate-nitrite due to low nitrate concentrations). SRP using an automated, high throughput, water chemistry machine, the Easychem Plus (Systea, Italy). All chemicals used to make standards and reagents were of reagent grade and methods used by machine standard. The unfiltered water samples were thawed then processed for total phosphorus (TP) using standard colourmetric methods and testing for absorbance on a Trilogy Laboratory Fluorometer using a PO4 module.
American Public Health Association (APHA), 1995
Soluble reactive phosphorus (SRP) µg/L
Total phosphorus (TP) µg/L Wetzel et al. 2000
Algal cover* % Visual estimate of % of the stream bed covered by algae.
Predictor variables
Macrophyte cover* % Visual estimation of % of the stream bed covered by macrophytes. Harding et al. 2009 Shading* % Visual estimate of % of the stream bed shaded when the sun is overhead.
Discharge m³/s
Discharge measured across one run, evenly flowing and free of obstructions. Offsets were placed wherever depth or discharge changed noticeably, with no fewer than five per transect. Water depth read on the downstream side of a ruler and water velocity is measured four-tenths of water depth up from the bed using a Marsh-McBirney Flow Mate. Discharge was calculated based on standard methods.
Gordon et al. 2004,
Harding et al. 2009
Wetted Width m Wetted width was measured at five locations along the 20m reach. Harding et al. 2009 * Variables considered constant, only assessed on the March sampling run.
Chapter Three 59
3.2.4 Land use, riparian conditions and other GIS derived variables
For each site, catchment area was defined using the River Environment Classification (REC, Snelder et
al. 2005). Due to the low topography of the Plains, catchments generated from digital elevations
only approximate actual catchment areas, so manual changes were made to fit known water
movement through ground truthing. This was further complicated by the stock water-races and
irrigation canals, which result in considerable cross-catchment transfer of water. To estimate the
relative inputs of these artificial waterways between sites, an upstream distance measure which
included their length was calculated. The elevation change across the reach that best approximated
the location of each site on the REC was calculated (Snelder et al. 2005), while stream order and the
number of junctions upstream were counted manually. All parameters related to stream channel
distance are based on the manually constructed stream maps. The area of each catchment situated
within a 100 m buffer of the stream channel was also calculated. For each catchment and 100 m
buffer zone area metrics of proportional land use, forest presence, urban development and road
density using ArcMap 9.2 tools, the New Zealand Land Cover Database ver.2 and AGRIBASE
(Terralink 2004 Agribase 2009 – Table 3).
Chapter Three 60Table 3: Distance, land use buffer zone and other site-wise variables derived by GIS, for use as predictors in analyses. Where no units are given the variable is a proportion. Site-wise Variables Methodology/calculation (all variables are based on the stream maps I constructed) UnitsStream distance to most downstream site The length along the stream channel to the most downstream site (Bramleys Rd) km Upstream natural channel length The length of all natural upstream channels (including all tributaries) based on traced and ground truthed
stream maps. The decision was made to “cut off” a natural stream reach (beginning the “unnatural” portion), when upstream reaches took the form of highly channelised, free-flowing canals, of uniform depth, characteristic of water-race or irrigation networks (Google Inc. 2009, ArcMap 9.2).
km
Upstream channel length including water race
The length of stream including all natural and man-made waterways upstream of the site. Where branches split in two in the direction of flow, I only measured the branch connecting to the system. (Waimakariri District Council water race data in geospatial form and ground truthing).
km
Upstream junctions Counted manually (excluding the water race network). Stream order Counted manually (excluding the water race network). Catchment area upstream Catchment areas were defined using the River Environments Classification (REC), then altered manually
to meet known directions of water flow and channels not delineated by this model (Snelder et al. 2005). km2
Area of 100-m buffer zone upstream The area upstream of each site within 100m of the natural stream. This distance was chosen based on its use in buffer zone delineation (Baker et al. 2007) and as the smallest distance class to measure land use metrics to assess critical distances of impact (Houlahan and Findlay, 2004) (ArcMap 9.2).
km2
Length of road per km2 of upstream catchment area
The total length of all paved and meteled roads in each upstream catchment area Zealand Landcover Database ver.2, (Terralink 2004), expressed as a fraction of the area in km2.
Proportion upstream catchment area that is urban
The proportion of the catchment area that is built-up (New Zealand Landcover Database ver.2, Terralink 2004))
Proportion upstream buffer zone that is urban
As above but for the 100-m buffer zone.
Proportion upstream catchment that is in moderate to high intensity farming
The proportion catchment area that is in use as Dairy, Beef, Sheep or Sheep and Beef (Agribase 2009)
Proportion upstream buffer zone that is in moderate to high intensity farming
As above but for the 100-m buffer zone.
Proportion upstream in each of each of 6 land use types
The proportion catchment area that is in use as dairy, sheep, beef, sheep and beef, arable and lifestyle as defined by the AGRIBASE (Agribase, 2009)
Proportion upstream buffer zone in each of each of 6 land use types
As above but for the 100 m near zone.
Proportion upstream catchment with forest cover
The proportion of the catchment covered in exotic or native forest (New Zealand Landcover Database ver.2, (Terralink 2004))
Proportion upstream buffer zone with forest cover
As above but for the 100-m buffer zone.
Elevation change within catchment. Based on the reach of highest and lowest elevation in all mapped stream reaches that correspond to REC reaches (Snelder et al. 2005).
m
Average level of macrophyte cover of a upstream sites
An average of the percentage cover of macrophytes of all upstream sites (Table 1) %
Average level of shading cover of a upstream sites
An average of the percentage cover of shading of all upstream sites (Table 1) %
Chapter Three 61
3.2.5 Analyses
3.2.5.1 Space-time interaction
I tested the influence of the space-time interaction on the composition of water quality in the Cam
River to determine if the spatial pattern varied through time and adapted the subsequent analysis
approach depending on the presence of a significant interaction (Legendre et al. 2010). A significant
space-time interaction would have led to separate analyses for each sampling period, whereas a non-
significant interaction meant subsequent analysis was able to be restricted to analysis of one sampling
period because the results obtained from different periods should be qualitatively similar (Laliberté et
al. 2009).
Canonical redundancy analysis (RDA) can be used as a form of multivariate analysis of
variance (MANOVA, manovaRDa) to test the relationship between a response matrix, typically species
abundance and two crossed factors (Legendre and Anderson, 1999). In this case, space and time were
the two factors and the water chemistry parameters were the equivalent of “species”. Orthogonal
dummy variables (Draper and Smith 1981), (otherwise known as Helmert contrasts were used to code
for the two factors. However, where no replication exists, as in this study, the space-time interaction
cannot be tested in the classic two-way ANOVA as no degrees of freedom remain to test the
denominator of the F statistic (Legendre et al. 2010). To avoid this problem, spatial and temporal
principal coordinates of neighbor matrices (PCNM) can be used to code for the space-time interaction
(Legendre et al. 2010). The PCNM method takes into account close neighbourhood relationships
among sites by using eigenvalue decomposition of a truncated matrix of geographic distances among
the sampling sites. The eigenvectors produced, corresponding to positive eigenvalues, are used as
spatial descriptors in regression or canonical analysis (Borcard and Legendre 2002, Dray et al. 2006).
The interaction is modelled using variables that are the product of the first s/2 and t/2 spatial and
Chapter Three 62
temporal PCNM variables, respectively (where s is the number of sites and t the number of sampling
times) (Laliberté et al. 2009). Empirical simulations show that this model has correct Type 1 error and
that its power is equal or greater than other potential ANOVA models (Legendre et al. 2010).
I used a mixed model in which space was considered a random factor and time fixed. Space was
considered random because although our site choices conformed to the selection protocol previously
mentioned, the path of the river across the landscape itself can be considered random. Time was
considered fixed as sampling occurred at regular intervals during the specific period of interest, in
which no major disturbance events occurred. A non-significant interaction effect led to the testing of
space and time using a basic RDA, without replication, rather than being treated as co-variables, as in
the previous analysis (Laliberté et al. 2009). All tests were performed with 999 permutations of the
residuals (Anderson and Legendre 1999). The manovaRDa procedure was carried out in the R program
(R Development Core Team 2007) using the package “STI”, with Model 5 used when the interaction
effect was included and Model 2 when without the interaction (Borcard and Legendre. 2002, Dray et al.
2006, Legendre et al. 2010 ) The “STI” contains functions found in both the ‘vegan’ and ‘PCNM’
packages (Dray et al. 2006, Oksanen et al. 2007).
3.2.5.2 Water quality change through time
I used non-multi-dimensional scaling (NMDS), a type of indirect gradient analysis that reduces the
dimensionality of multivariate data to aid in the interpretation of water quality in the Cam River
(Jongman et al. 1995). NMDS has been shown to be a more robust method of indirect gradient analysis
than principal components analysis or detrended correspondence analysis (Minchin, 1987). NMDS
reduces the dimensionality of multivariate data by describing major trends among sites by the joint
occurrence of similar levels of each parameter or species (Jongman et al. 1995). Much variation is often
Chapter Three 63
captured in two or three derived axes which can then be related graphically or statistically to other
variables to reveal patterns (Hawkins et al. 1997).
Euclidean distance is commonly used when conducting all types of gradient analyses on
environmental data (Clarke and Corley 2006). However, Euclidean distance is strongly influenced by
whether the data are rescaled or standardised in ordination procedures (Jackson 1993). The scaling of
variables is important when dealing with environmental data, as they often have different units of
measurement meaning that large numbers will completely obscure variation in the small numbers,
resulting in a matrix that is dominated by one variable. This problem was avoided by standardising the
variables so the each variable was given an equal weight (McCune et al. 2002). I did this using
normalisation in PRIMER, which involved subtracting the mean and dividing by the standard deviation
for that variable (Clarke and Corley 2006). It sets the mean at 0 and standard deviation at 1 and each
variable is adjusted to an equivalent value in the range of approximately -2 to 2. This achieves a local
shift and scale change crucial for making such variables comparable with each other (Clarke and
Corley 2006). However, this procedure does not solve problems related to non-normality so
transformations were done first. Phosphorus variables (SRP and TP) were log-transformed, while DIN
and turbidity were square-root-transformed, as these variables were less skewed.
Non-metric multidimensional Scaling (NMDS) using Euclidean distance and 999 permutations
was carried out on the transformed set of water quality data for all sites and samples over the three
repeats in 2010 and 2009 sampling of the Cam River in Primer (Clarke and Corley 2006). This resulted
in a comparable measure of water quality across all sampling runs. Euclidean distance in NMDS space,
between sites, across the three sample runs, was calculated and used to measure change in water quality
variables between sampling occasions (as in Scarsbrook 2002).
To assess if the change in water quality was greater between any two consecutive sampling runs
compared with any other two consecutive runs, paired-sample T tests were performed on the change in
Chapter Three 64
NMDS space values between corresponding sampling occasions (Zar 1999, R Development Core Team
2007).
3.2.5.3 Factors relating to change in water quality composition
I examined the relationship between the amount of change in water quality that occurred at sites
between the January and February sampling events and the size of sites using Quantile regression. The
plot of change of NMDS space against the log of each size variable gave a better spread to the data
with a more observably linear decrease in NMDS change, so was used in analysis. Quantile regression
was performed to fit regression lines that bound the upper 10 percent of the data in the R package
quantreg using the "br" method, which is a variant of the Barrodale and Roberts (1974) simplex
algorithm and is suitable for data sets of this size (R Development Core Team 2007, Koenker 2009).
These were tested for significance using the "boot" method to compute standard errors (Koenker 2009).
To determine which landscape level factors influenced the amount of change in water quality
composition through time I used a model selection approach in the program Spatial Analysis in
Macroecology (SAM, Rangel et al. 2006). This procedure uses Akaike’s Information Criterion (AICc)
to select the best model out of many competing models providing a parsimonious balance between
model complexity and predictive power (Rangel et al. 2006, Rangel et al. 2010). SAM evaluates all
models that emerge from a set of explanatory variables and ranks them according to their AICc value
and derived statistics (e.g. Akaike’s weight and delta AICc). A single predictive “multi model” can be
generated by averaging and weighting the estimated model parameters as a function of Akaike weights
(Rangel et al. 2010). Although this model is based on ordinary least squares regression, the problem of
inflated type one errors (due to spatial autocorrelation in the residuals) can be tackled by adding spatial
covariates as “fixed” predictors in the model selection procedure (Diniz-Filho et al. 2008). I used
“stream distance to most downstream site” as a fixed predictor in the model selection process as many
Chapter Three 65
other predictors were related to flow and this measure is more meaningful than Euclidean distance in
the case of streams (Lyon et al. 2008). Results are displayed for models with a change in AICc (∆AICc)
of less than 2, as these are statistically equivalent to the minimum AICc model (Rangel et al. 2006).
3.2.5.4 The impact of tributaries on longitudinal water quality variation
To initially explore system-wide patterns in water quality, I displayed each parameter visually using
graduated colour symbols to represent the value of each site, at its actual location. I studied longitudinal
changes in the instantaneous flux of nutrients (total phosphorus, soluble reactive phosphorus and DIN)
along the Northbrook, the longest branch of the Cam River, using a mass balance approach. Based on
the results of the space-time interaction test, analysis of the spatial patterns analysis of water quality
from one sampling event is sufficient (Laliberté et al. 2008) Thus this approach was restricted to the
March snapshot sampling event as this was the most intensive.
Total instantaneous loads of each nutrient species were calculated for each site along the
mainstem of the Northbrook and the site closest to the confluence on each tributary, by multiplying the
concentration by the discharge at that site and time (similar to the method of Grayson et al. 1997). It is
common in water quality studies for point-sources total loads (tributary and drain inputs in this case) to
be subtracted from the total loads at main stem sites in a longitudinal manner (Grayson et al. 1997,
Salvia et al. 1999, Behrendt and Opitz 1999). If the remainder is a positive number, the increase is
attributed to non-point source inputs along the intervening reach (Grayson et al. 1997), whereas a
negative number is attributed to in-stream uptake by biota or sedimentation (Finlay et al. 2011, Salvia
et al. 1999). This mass balance approach was used to calculate expected values of total and dissolved
phosphorus, DIN for sites along the Northbrook branch, and then compared to the actual instantaneous
flux. I calculated the deviation difference between observed and expected values at each site along the
main stem of the Northbrook. I also plotted the observed and expected flux of each nutrient in relation
Chapter Three 66
to the distance of sites from the source of the Northbrook. The flux of each nutrient at tributary sites
was superimposed at the distance at which they entered the mainstem, to allow estimation of their
influence.
3.3 RESULTS
3.3.1 Space-time interaction
The space-time interaction on water quality was not significant when comparing between the 2010
sampling events or between the March 2009 and March 2010 events (P =0.40 and 0.63, respectively,
Table 4). Thus the spatial patterns of water quality were stable through time and temporal variations
were common to all sites. Tests of the main effects of space and time showed that water quality in the
Cam River was strongly spatially structured for both combinations of sampling events tested (P
=0.001); this was true irrespective of the exclusion or inclusion of the interaction term in the model for
the test of 2010 events (Table 4). By comparison, water quality did not vary significantly in time, in
either combination of sampling events considered, when the interaction term was excluded from the
model (P = 0.8 and 0.27 respectively, Table 4). This means that there were few temporal changes in
water quality over the 2010 summer and between the months of March in each summer.
Chapter Three 67
Table 4: Effects of space, time and their interaction on the levels of all physico-chemical water quality measures during
repeat sampling events of the Cam River. Spatial and temporal principal coordinates of neighbour matrices (PCNM –
Borcard and Legendre 2002) were used to code for the space-time interaction in a manova-like RDA procedure (Laliberté et
al. 2009). A mixed model was used in which space was considered a random factor and time fixed. A non-significant
interaction led to the testing of space and time using a basic RDA, without replication. Analyses were conducted using the
“STI” package in the R program (R Development Core Team 2007).
Three sample events in 2010 March 2009 to March 2010
R² Fª P R² Fª P
Source
Space x time _ 1.12 0.40 _ 0.81 0.63
Interaction excluded in the model
Space 0.52 2.28 0.001* 0.56 1.29 0.001*
Time 0.008 1.12 0.8 0.011 1.07 0.27*
Interaction included in the model
Space 0.52 2.35 0.001* 0.56 1.17 0.28
Time 0.008 1.10 0.21 0.011 1.19 0.21
* Values are significant at P ≤ 0.05. All tests were performed using 999 permutations of the raw data.
ª Pseudo-F
3.3.2 Water quality changes through time
Eight measured physico-chemical variables were included in the NMDS analysis. Axis one of the
NMDS was strongly correlated (Spearman rank correlation rs > |0.6|) with total phosphorus and
turbidity (rs = -0.75 and -0.69 respectively). While axis two was strongly correlated with nitrate (rs = -
0.63) and dissolved oxygen (rs = -0.69) (R Development Core Team 2007). Many sites remain central
and move only a small distance in NMDS ordination space between sampling events, however, there
were some sites with more extreme values in each sampling event (Figure 2).
No significant differences were found between the consecutive levels of change in water quality
across the four sampling occasions using paired two-sample t tests (January to February compared with
February to March, p = 0.095; March 2009 to Jan 2010 compared with January to February 2010, p =
Chapter Three 68
0.097). This suggests that the amount of change in water quality was approximately equal between
each sample run.
Axis 1
-2 0 2 4
Axis
2
-2
-1
0
1
2March 2010 Feburary 2010 January 2010March 2009
Stress = 0.16
Figure 2: Water quality of sites sampled on the Cam River in each of four sampling runs represented in two dimensional
Non-metric multidimensional scaling (NMDS) ordination space. A stress value of less than 0.2 and indicates a good
representation of community relationships between sites.
3.3.3 Factors relating to change in water quality composition through time
Sites on smaller sections of the river, lower in discharge and closer to headwaters, had greater changes
in water quality (Between the January and February sampling events) than larger sites further down the
network (Figure 3). Smaller sites were also more variable in terms of the amount of change in NMDS
space that occurred between sampling runs (Figure 3). The upper level of variation in the temporal
change in water quality significantly reduced with increasing discharge and length of upstream natural
Chapter Three 69
channel (p = 0.004 and p ≤ 0.001, Figure 3. a, and b, respectively). The temporal variation of water
quality in small streams was highly variable when compared with the homogeneous nature of larger
streams.
The best model (determined by Akaike’s Information Criterion, AICc), explained 20% of the
variance in the relationship between catchment, buffer and in-stream variables and the amount of
change of water quality between sampling events. The variables stream order, catchment area upstream
and the proportion of the catchment area in dairy and forested were included in the best model (Table
4). However 17 models performed equally well (∆AICc < 2) and included combinations of 14 variables,
(including stream distance to lowest site, which was held constant) of the many which were put
forward for model selection (All from Table 3 and predictors from Table 2). The four most important
variables (averaged over all models) explaining change in water quality were stream order, the
proportion of the catchment area in dairy and forested, and upstream channel length including water-
race (listed in order of decreasing importance, Table 5). The change in water quality levels was greater
in reaches that were smaller and further up in the system, and also greater in those with a higher
proportion of catchment area that was forested and a lower proportion as dairy farm (relative to other
sites on the network) (Coefficients, Table 5)
Chapter Three 70
0
1
2
3
4
Cha
nge
in N
MD
S o
rdin
atio
n sp
ace
Figure 3: Change in Non-metric multidimensional
scaling (NMDS) ordination space between sites in
the January and February sampling events against the
average discharge across both sampling events (a)
and the length of natural stream channel upstream of
a site (b). The upper 90th Quantile regression lines are
shown on each. Note the log scale on the x axis of
both graphs.
0.001 0.01 0.1 1
Table 4: Parameter estimates for variables selected in the best Ordinary Least Squares model according to the Akaike
Information Criterion (AICc) after testing combinations of all possible models. This model explained the most variance, as
determined by the R² and AICc, in the amount of change in water quality (community of physico-chemical variables),
between the 2010 January and February sampling events in the Cam River. The response variable was the average change in
Euclidean distance in the water quality Non-metric Multi Dimensional Scaling (NMDS) ordination space. The predictor
variables were the suite of catchment, buffer and in-stream variables identified as predictors in Table 1 and all those in
Table 2, in addition to the spatial autocorrelation variable held fixed. AICc indicates the fit of each model, taking into
account the number of variables to avoid over fitting.
Variable Coefficient Standard Error Constant 1.82 0.38
Stream order -0.45 0.12 Catchment area upstream <0.001 <0.001 Proportion of upstream catchment area in dairy -0.96 0.38
Proportion of upstream catchment area forested 5.92 2.98
Stream distance to most downstream site -0.073 0.046 Note that “Stream channel distance to lowest site” is included in every model to account for the spatial location of each site.
Discharge (m3s-1)
Length of ustream natural stream channel (km)
0.1 1 10 1000
1
2
3
4
(a)
(b)
Chapter Three 71
Table 5: The importance (the number of times the variable was selected, averaged over all models) of catchment, buffer and
in-stream variables predicting change in water quality composition between the January and February sampling events in
2010, using the Akaike Information Criterion (AICc) model selection approach. Variables reported were chosen at least
once in models that had a change in AICc (∆AICc) of less than 2, indicating that they are not significantly different in their
ability to predict change in the water quality composition. The number of time each variable was selected in these models
was indicated. The spatial variable, “stream distance to most downstream site” had an importance of one as it was included
in the selection as a fixed predictor, thus was forced to be included in every model. The coefficients are shown to indicate
the direction of the relationships.
Variable Importance No. times selected Coefficient Standard
Error Stream order 0.78 16 -0.37 0.13 Proportion of upstream catchment area in dairy 0.70 14 -1.08 0.40 Proportion of upstream catchment area forested 0.57 14 5.89 1.99 Upstream channel length including water race 0.47 2 -0.007 0.002 Proportion of upstream catchment area in sheep 0.46 2 3.46 1.24 Proportion of upstream catchment area lifestyle 0.43 2 -2.04 0.68 Catchment area upstream 0.39 5 <0.001 <0.001 Proportion of upstream catchment area urban 0.37 6 -0.50 0.22 Area of 100 m buffer zone upstream 0.31 2 <0.001 <0.001 Proportion of upstream catchment area in beef 0.30 1 -1.61 0.58 Wetted width 0.25 2 0.06 0.03 Upstream natural channel length 0.25 1 0.004 0.006 Average discharge 0.24 1 0.29 0.18 Stream distance to most downstream site 1.00 17 -0.05 0.06
3.3.4 The impact of tributaries on longitudinal water quality variation
Sites along the Northbrook branch of the Cam River system cover a wide range of each of the nutrient
parameters measured (Figure 4, Table 6). The instantaneous flux of total phosphorus (TP), soluble
relative phosphorus (SRP) and dissolved nitrate-nitrite (DIN) increased with increasing stream channel
distance from the source of flow (Table 4, Figure 6). The flux in the lower reaches of the Cam River
was greater than that in the upper Northbrook by one order of magnitude for TP and SRP and two
orders of magnitude for DIN. Sharp changes in the observed fluxes were predicted well by the expected
Chapter Three 72
fluxes when responding to the entry of tributaries. The magnitude of these changes was generally
proportional to the flux of nutrients in incoming tributaries (Figure 6).
The flux of TP and SRP was less than expected in the lower reaches of the Cam River and
began to decline from approximately 5-6 km downstream from the source of flow. At this point the
large tributaries, the Southbrook and Coldstream had combined with the Northbrook to form the main
Cam River. There were mismatches between the observed and expected values through the mid-section
of the Northbrook in the flux of both TP and SRP, the expected values variously over and under
estimated the observed levels, with a pattern that was not consistent between the forms of phosphorus.
The observed flux of DIN matched the expected flux more closely than those of TP and SRP. At 5-6
km downstream of the source, following the confluence with the Southbrook and Coldstream, the flux
of DIN in the Cam River began to decrease, however this decrease was only gradual compared with a
steep decline in the expected values of DIN flux. This resulted in the expected values of DIN flux
underestimating the observed levels in the lower reaches of the Cam River.
Table 6: Mass balance of the instantaneous flux of total phosphorus, phosphorus and dissolve nitrate and nitrogen (DIN) at
sites along the Northbrook branch and then mainstem of the Cam River during the March sampling event in 2010. Each row
contains data from one site, sites become further from the headwaters moving down the table, rows containing mainstem
sites are shaded. The “Mainstem” and “Tributary” columns give the flux based on water sample nutrient concentrations and
discharge of this sampling event, at all sites along the main branch of the river and at the site on each tributary closest to the
confluence with the mainstem, respectively. The values in the “Expected mainstem” columns represent the flux expected at
a site based on the flux at the site directly upstream and with the addition of any tributary flux(es) that have entered. The
“Diff from expected” column gives the difference between observed and expected flux at each mainstem site, the four
largest differences in each case are in bold
Chapter Three 73
Sites Instantaneous total phosphorus flux (ug/s) Instantaneous SRP flux (ug/s) Instantaneous DIN flux (mg/s) Downstream
grasslands (Hawkins 1957). Several mills profited from the water in swamps and small creeks that
drained into the Cam River. These produced a variety of wastes, many of which were disposed of
into the river. Agriculture and farming became the primary industries after water from the swamps
was drained and soils became more productive (Hawkins 1957). In more recent years, the non-urban
parts of the catchment have become dominated by dairy farming and lifestyle blocks (Biggs 1985).
All waterways flowing into the Cam River system were mapped; this included all drains,
channelised stream segments and water races or irrigation canals that entered the system. A GIS
layer of all water flowing into the Cam River was built by tracing the river lines using Google Earth
images (the most accurate and high quality images available), in combination with field knowledge
and ground truthing, using a GPS (Garmin GP560) (Figure 1). Sixty-one sites were spread
throughout the Cam River system as per the “snapshot” methodology (Chapter 1) (algal production
was only measured at a subset of these sites). With this method I aimed to produce an instantaneous
picture of all concentrations and fluxes in the watershed, by sampling every confluence and
discharge point within a short time period during base flow (Grayson et al. 1997). For ecosystem
processes this allowed coverage of all parts of the system where conditions may potentially change.
Sites were placed, where feasible on all first-order tributaries and point sources entering the system.
They were situated at least 25 m above and below every point of confluence, so that no two sites
were less than 50 m apart. Sites were also placed along uninterrupted main stem and tributary
reaches, approximately 500 m apart (Figure 1).
Chapter Four 97
Figure 1: The location of the Cam River on the Canterbury Plains (a) and the location of the Cam River sampling sites
(b). The blue lines show the natural channels of the Cam River system (constructed by tracing from Google Earth
images and ground truthing). The stream drains towards the bottom right. The river lines and sampling sites are overlain
on a topographical map of the area. The nearby town Rangiora and three major branches of the river are identified, these
branches confluence in quick succession to the south of the town, to form the main Cam River.
4.2.2 Measuring ecosystem processes
4.2.2.1 Algal productivity
I used levels of chlorophyll a in caged, unglazed ceramic tiles (colonisation surfaces) as relative
measures of algal growth between the sites. The tiles (5 x 5 cm2) were positioned inside cages (250
ml plastic containers with four, 5 x 5 cm2 mesh panels) which were glued inside polystyrene floats
(rings, 10 cm inner radius, 15 cm outer radius and 3 cm deep), attached with a cord to a metal stake
in the streambed. In this way the tiles at all sites remained at 10 cm depth below the surface and
were able to sway in the current, deflecting debris and were not accessible to grazing invertebrates.
Water velocity in the containers was reduced, but proportional to the surface current velocity. The
intent of placing the tiles in the floats was to prevent insect colonisation and consumption of the
Chapter Four 98
algae, so that the algal growth would primarily reflect the water conditions in the stream. At each
site, three replicate floats, with 1 tile in each, were placed every 5 m, at random offsets from the
stream bank. At each float, the percentage of overhead shade cover was estimated visually, water
depth read on the downstream side of a ruler and water velocity measured, using a Marsh-McBirney
Flow Mate, at the stake (at 4/10ths of the water depth) and the float (directly upstream of a window
in the plastic container, to measure velocity across the tile). After three weeks exposure (late
February to March) the tiles were removed, kept on ice in the dark and then frozen.
The chlorophyll was extracted in 70 ml of unbuffered 90 % ethanol over 24 hours at 4 ºC
(adapted from Biggs and Kilroy 2000). Three subsamples of 3 ml of extract were gently removed, so
as not to disturb sediments, transferred to a 1-cm glass cuvette and the absorbance was read at
wavelengths, 665 nm and 750 nm (turbidity correction), using a uv-visible spectrophotometer
(Shimadzu, PharmaSpec uv-1700). The extract was then acidified with 0.3 ml of 0.1N HCl (to
correct for the measurement of pheopigments) and absorbance re-read at the previous wavelengths
(Biggs and Kilroy 2000). Chlorophyll a was calculated in mg-1m2 using the absorption coefficient
for chlorophyll a as defined by Sartory and Grobbelaar (1984) and averaged across tile extract
replicates and within each site.
4.2.2.2 Leaf decomposition and microbial respiration
I used packs of willow leaves to measure variation in the rate of leaf breakdown and microbial
respiration on leaves, between sites, on the Cam River. Willow (Salix spp.) leaves were collected the
previous autumn from the University of Canterbury campus and air dried completely. Each pack
consisted of a 5 g (±0.1 g) portion of willow leaves (weighed on an analytical balance) packed
inside a fine mesh onion bag (1 mm mesh) (Benfield 2006). Three packs per site were attached
individually to metal stakes, driven into the streambed every 5 m, at random offsets from the stream
bank (algal tile floats attached to the same stakes). Areas likely to be unstable or eroded were
Chapter Four 99
avoided. The packs were attached so as to float just above the stream bottom, downstream of the
stake to prevent accumulation of debris. The water depth at each pack was read on the downstream
side of a ruler and water velocity measured using a Marsh-McBirney Flow Mate (at 4/10ths of the
water depth). The leaf packs were collected after four weeks, and transported on ice to the
laboratory. At the time of collection, 500 ml of filtered stream water was collected from each site in
acid washed plastic bottles and transported on ice to the laboratory. Within 2 hours of collection,
each pack was opened and 3 leaves (or partial leaves where no whole leaves remained) were
removed in order to study microbial respiration rates on leaf material between sites.
Oxygen consumption was measured on the cores taken from the leaf litter as a measure of
microbial activity on the decomposing litter (methods adapted from Niyogi et al. 2001). The filtered
stream water was kept at 15 ºC and oxygenated to the point of saturation (at least 30 minutes), using
a central air system prior to use. The leaves were gently rinsed in filtered stream water then two 1
cm diameter cores were removed from each leaf with a metal corer and transferred to a 35 ml plastic
vial (autoclaved previously at 120 ºC for 15 minutes for sterilisation). As not all packs contained
enough complete leaves to take full cores, in some instances the material consisted of several partial
cores which approximated the surface area of two complete cores. Vials were then filled with
saturated, filtered stream water from the corresponding site and covered tightly with parafilm,
avoiding air bubbles. The incubations were performed in 15ºC water-baths to prevent diffusion of
oxygen into the vials through the plastic, and lasted 30-40 hours. Niyogi et al. (2001) found that
oxygen uptake on willow leaves was linear during the course of incubations. Vials were gently
agitated during the incubation to prevent the formation of a diffusion gradient (Niyogi et al. 2001).
Additional treatments of vials containing only saturated, filtered stream water were run as controls
to ensure that oxygen uptake was not significant compared with that of litter samples (Niyogi et al.
2001). At the end of the incubation period, oxygen levels in each vial were measured (YSI 550
dissolved oxygen meter) and the leaf material from each vial dried and weighed. Microbial
Chapter Four 100
respiration was averaged within packs and sites and calculated as the micrograms of oxygen
consumed per milligram of leaf matter per hour.
The amount of leaf litter break-down that had occurred at each site was calculated from the
change in leaf mass during the treatment. Leaf litter from each pack was passed twice through a 2.5
mm sieve, washing gently, to remove small invertebrates and all forms of fine particulate matter.
Macroinvertebrates, stones, sticks, leaves of different species and any other material obviously
foreign to the litter pack were removed and the remaining willow leaves dried at 50 ºC for at least
24 hours (Benfield 2006). As many packs contained a large amount of fine sediment, the leaves
were weighed, then combusted in an ashing oven (McGregor Muffler, 550 ºC for 2 hours) and
reweighed to calculate the ash free dry mass (AFDM) of the leaf material. The difference in weight,
before and after ashing, was used to calculate the mass of sediment that was attached to the leaves;
this was divided by the AFDM to give a proportional weight of sediment accumulation.
4.2.2.3 Trophic state
Considerable additional information about the trophic status of a river can be obtained from the ratio
of nitrogen to phosphorus in stream water. Spatial variation in the trophic state throughout the Cam
River was studied using three snapshot samples of the water quality (Grayson et al. 1997). There is
disagreement about which measures should be used to form the ratio (Grayson et al. 1997), but in
this study I use the ratio of dissolved nitrate and nitrate (assumed as equivalent to DIN, due to the
small proportion that was nitrite) to dissolved phosphorus (SRP), as these are the major forms of
nutrients available for algal growth. It has been suggested that this is the best representation of the
trophic status, as experienced by productive biota. The methods associated with the collection of the
snapshot water quality data will be described in the following section. The interpretation of the
trophic state of the Cam River was based on the assumption that for algal growth the ratio should be
Chapter Four 101
16:1 mol of N to P. A ratio of less than 10 indicates nitrogen limitation and a ratio of greater than
20, phosphorus deficiency (Stelzer and Lamberti 2001).
4.2.3 Predictor variables
4.2.3.1 Physico-chemical water quality
Water samples were collected in two events (February and March), spread over 1 to 4 days, at the
start and end of the deployment of the algal tiles and leaf packs. To meet the requirements of
snapshot sampling (Chapter 1), sampling was conducted within a period of base flow, defined as the
period between storms when the hydrograph was in the later stages of the recession limb, considered
as at least two days after a storm flow peak (Pionke et al. 1999). Sampling was not conducted if rain
had occurred in the previous week in the catchment, Southern Alps or foothills which may cause the
Waimakariri and Ashley Rivers to rise, thereby raising the water table feeding the springs in the
Cam River. In light of the relatively small variation in discharge across each sampling period, I
considered flow in the Cam River to be stable during each sampling period. Sampling was only
conducted up to 5 hours either side of midday to minimise diurnal variation in parameters such as
dissolved oxygen and temperature. Diurnal variation in water chemistry is minimal with regards to
all major forms of nitrogen and phosphorus except for ammonia (not analysed here) (Finlay et al.
2011). I measured eight physical and chemical properties of the water, as well as discharge, at each
visit to a site (field and laboratory processes described in Table 1). Each day I visited downstream
sites first and collected water samples immediately on arrival at a site to avoid contamination from
sediments disturbed by entering the water. The average values of the sampling events were used
when analysing, as this reflected the overall conditions each measure was exposed to.
Chapter Four 102
4.2.3.2 Other in-stream variables
Other in-stream and site properties, which were considered more permanent were assessed on only
one occasion (methodology in Table 1). Macrophyte cover, algal cover and the proportion of the site
shaded were estimated visually (Table 1). Stream size-related variables include stream bank full
width, stream order and discharge. Although discharge was measured during each visit to a site,
average discharge was used in analyses as this better described the conditions during each measure
of ecosystem processes. Water temperature loggers (HOBO pendant temperature/light, Onset
Corporation) were installed underwater, just above the streambed at a subset of sites, covering the
period of algal tile and leaf pack deployment. Data was logged every 30 minutes. Data from the
three week period when the algal tiles and leaf packs were in the stream was averaged (5 hours
either side of midday) for consistency with spot measurements. As average spot temperatures during
the site visits correlated well with average temperatures from the loggers, I estimated the
temperature at sites without loggers by averaging the spot temperatures.
Chapter Four 103Table 1: Field and laboratory methodology for the eight water quality variables and other predictor variable measured at each site, as well as treatment in analyses.
Variable Units Field and/or laboratory methodology Reference Treatment Temperature ºC Time averaged spot measurement. The YSI Sonde 6600, a multi-parameter, water quality
measurement device was placed in the thalweg of a run at the top of each site, continuously logging during time spent at each site. The final 5-20 minutes of data from each site was averaged (visit times varied depending on sample run intensity).
I collected 80 mL of water in a syringe from the thalweg of the stream just below the surface and pushed through a filter (Whatman GFF 250 Millipore rating) into a 100mL opaque plastic bottle (pre-soaked in 5% hydrochloric acid overnight then rinsed three times with distilled water and a further three times with milli-q water). 80 mL of unfiltered water was collected in a similar way. Samples were kept on ice and then frozen. Filtered samples were thawed then analysed colourmetrically for DIN (assumed from combined nitrate-nitrite due to low nitrate concentrations). SRP using an automated, high throughput, water chemistry machine, the Easychem Plus (Systea, Italy). All chemicals used to make standards and reagents were of reagent grade and methods used by machine standard. The unfiltered water samples were thawed then processed for total phosphorus (TP) using standard colourmetric methods and testing for absorbance on a Trilogy Laboratory Fluorometer using a PO4 module.
American Public Health Association (APHA), 1995
Soluble reactive phosphorus (SRP) µg/L
Total phosphorus (TP) µg/L Wetzel et al. 2000
Algal cover* % Visual estimate of % of the stream bed covered by algae.
Predictor variables
Macrophyte cover* % Visual estimation of % of the stream bed covered by macrophytes. Harding et al. 2009 Shading* % Visual estimate of % of the stream bed shaded when the sun is overhead.
Discharge m³/s
Discharge measured across one run, evenly flowing and free of obstructions. Offsets were placed wherever depth or discharge changed noticeably, with no fewer than five per transect. Water depth read on the downstream side of a ruler and water velocity is measured four-tenths of water depth up from the bed using a Marsh-McBirney Flow Mate. Discharge was calculated based on standard methods.
Gordon et al. 2004,
Harding et al. 2009
Wetted Width m Wetted width was measured at five locations along the 20m reach. Harding et al. 2009 * Variables considered constant, only assessed on the March sampling run.
.
Chapter Four 104
4.2.3.3 Land use, riparian conditions and other GIS derived variables
For each site, catchment area was defined using the River Environments Classification (REC), a
GIS-derived database of river and stream networks in New Zealand (Snelder et al. 2005). Due to the
low topography of the Plains, catchments generated from digital elevations only approximate actual
catchment areas, so manual changes were made to fit known water movement established through
ground truthing. This process was further complicated by the stock water-races and irrigation canals,
which resulted in considerable cross-catchment transfer of water. To estimate the relative inputs of
these artificial waterways I used an upstream distance measure which included the length of the
water channels races (not sampled) that entered the system. All parameters related to stream distance
were based on the stream map constructed from Google Earth images (Google Inc. 2009) and field
knowledge. All parameters relating to Euclidean distance are based on the Cartesian Co-ordinates of
the sites. For each site, measures of land use, riparian conditions, road density and in-stream
conditions at upstream sites were derived for the catchment area contributing to the site and the area
upstream area 100 m either side of the stream, using ArcMap 9.2 tools and the New Zealand Land
Cover Database ver.2 (Terralink 2004 – Table 2).
Chapter Four 105Table 2: Distance, land use buffer zone and other site-wise variables derived by GIS for use as predictor in analyses, where no units are given the variable is a proportion. Site-wise Variables Methodology/calculation (all variables are based on the stream maps I constructed) UnitsStream distance to most downstream site The length along the stream channel to the most downstream site (Bramleys Rd) km Upstream natural channel length The length of all natural upstream channels (including all tributaries) based on traced and ground truthed
stream maps. The decision was made to “cut off” a natural stream reach (beginning the “unnatural” portion), when upstream reaches took the form of highly channelised, free-flowing canals, of uniform depth, characteristic of water-race or irrigation networks (Google Inc. 2009, ArcMap 9.2).
km
Upstream channel length including water race
The length of stream including all natural and man-made waterways upstream of the site. Where branches split in two in the direction of flow, I only measured the branch connecting to the system. (Waimakariri District Council water race data in geospatial form and ground truthing).
km
Upstream junctions Counted manually (excluding the water race network). Stream order Counted manually (excluding the water race network). Catchment area upstream Catchment areas were defined using the River Environments Classification (REC), then altered manually
to meet known directions of water flow and channels not delineated by this model (Snelder et al. 2005). km2
Area of 100-m buffer zone upstream The area upstream of each site within 100m of the natural stream. This distance was chosen based on its use in buffer zone delineation (Baker et al. 2007) and as the smallest distance class to measure land use metrics to assess critical distances of impact (Houlahan and Findlay, 2004) (ArcMap 9.2).
km2
Length of road per km2 of upstream catchment area
The total length of all paved and meteled roads in each upstream catchment area Zealand Landcover Database ver.2, (Terralink 2004), expressed as a fraction of the area in km2.
Proportion upstream catchment area that is urban
The proportion of the catchment area that is built-up (New Zealand Landcover Database ver.2, Terralink 2004))
Proportion upstream buffer zone that is urban
As above but for the 100-m buffer zone.
Proportion upstream catchment that is in moderate to high intensity farming
The proportion catchment area that is in use as Dairy, Beef, Sheep or Sheep and Beef (Agribase 2009)
Proportion upstream buffer zone that is in moderate to high intensity farming
As above but for the 100-m buffer zone.
Proportion upstream in each of each of 6 land use types
The proportion catchment area that is in use as dairy, sheep, beef, sheep and beef, arable and lifestyle as defined by the AGRIBASE (Agribase, 2009)
Proportion upstream buffer zone in each of each of 6 land use types
As above but for the 100 m near zone.
Proportion upstream catchment with forest cover
The proportion of the catchment covered in exotic or native forest (New Zealand Landcover Database ver.2, (Terralink 2004))
Proportion upstream buffer zone with forest cover
As above but for the 100-m buffer zone.
Elevation change within catchment. Based on the reach of highest and lowest elevation in all mapped stream reaches that correspond to REC reaches (Snelder et al. 2005).
m
Average level of macrophyte cover of a upstream sites
An average of the percentage cover of macrophytes of all upstream sites (Table 1) %
Average level of shading cover of a upstream sites
An average of the percentage cover of shading of all upstream sites (Table 1) %
Chapter Four 106
4.2.4 Analyses
4.2.4.1 Patterns in ecosystem processes
To initially evaluate system-wide patterns in ecosystem processes, each measure of ecosystem
processes was expressed visually using graduated colour symbols to represent the value of each site, at
its actual location on a map of the streamline (as in Tu and Xia 2008, ArcMap 9.2). Correlation
between the ecosystem process measures was assessed using the non-parametric, Spearman’s rank
correlation procedure (two-tailed, α = 0.05), to overcome problems associated with lack of normality.
To test whether variability in each ecosystem process changed with stream size, I fitted linear
functions to the upper and lower limits of relationships between each ecosystem process and discharge
using quantile regression in R (‘quantreg’ package; Cade and Noon 2003, R Development Core Team
2007, Koenker 2009). Discharge, width, stream order and kilometres of stream channel upstream of a
site aligned along the same axis under principal components analysis, so discharge was a suitable
measure to generally represent stream size. I also tested if discharge was related to these limits by
investigating limit responses for each measure of ecosystem processes when related to stream size, by
regressing the sites that fell within upper (90th) or lower (10th) quantiles (depending on which
relationship was investigated) of the ecosystem process-size relationship against discharge. Confidence
intervals were computed by the rank inversion method and P-values by bootstrapping (Koenker 2009).
4.2.4.2 Partitioning the variation to space and environment
One way to estimate the relative contribution of the environment (water quality, land-use and in-stream
conditions) and other spatial processes to patterns in ecosystem processes, is to partition the variation
measured between environmental and spatial factors, using spatial eigenfunctions (Cottenie 2005,
Peres-Neto and Legendre 2010). This method is related to the method used in community ecology to
Chapter Four 107
separate the contribution of niche processes (environmental conditions) and other spatial processes
(such as dispersal) on community structure (Borcard et al. 1992, Legendre et al. 2005). Provided that
the environmental factors have been appropriately quantified (with relevant variables), the extent of
their influence can be quantified, and residual spatial variation attributed to other spatial processes,
such as flow of solutes and matter within the system. I have used this method to partition the variation
in ecosystem processes into four fractions; pure environmental (E|S), pure spatial (S|E), jointly
explained (environment-spatial) and unexplained variation (Legendre et al. 2005). It should be noted
that part of the environment-space fraction could be due to other spatial processes that show correlation
with the environment (Bell et al. 2006) and that the pure spatial fraction may hide the effects of some
unmeasured spatially structured environmental variables (Borcard and Legendre 1994, Jones et al.
2008).
I used Eigenfunction analysis to represent the spatial fraction. This involved constructing a
series of spatial eigenvectors using principal coordinates of neighbour matrices (PCNM) and related
methodologies (Borcard and Legendre 2002, Dray et al. 2006). These methods are based on, and
comparable, to the Moran’s I statistics, which are the most commonly used statistics for spatial
autocorrelation analysis (Rangel et al. 2006). The PCNM method codes spatial information in a way
that allows one to recover various structures over the whole range of scales that the sampling design
passes (Borcard and Legendre 2002). Empirical results show that the positive eigenvectors alone give a
good representation of the spatial relationships (Borcard and Legendre 2002, Borcard et al 2004).
These eigenvectors represent an orthogonal spatial structure (as they are the product of a symmetric
matrix) and can be used as explanatory variables in regression or canonical models (Dray et al. 2006).
Eigenvectors (constructed by any method) with large eigenvalues describe global structures, whereas
those with small eigenvalues describe local structures (Borcard and Legendre 2002).
Chapter Four 108
I used eigenfunction-based techniques to create a series of metrics to describe the spatial
structure in ecosystem processes, based on different distance measures. I then assessed how well each
metric accounted for variance in the ecosystem processes as a multivariate suite and each individually
(Griffith and Peres-Neto 2006, Blanchet et al. 2008b). I used both Euclidean distance (PCNM-E) and
stream channel distance (PCNM-S), to construct two sets of spatial eigenvectors using the basic PCNM
methods, to test the importance of overland, direct distances and in-stream distances, respectively, on
structuring ecosystem processes, considering all sites as connected (Table 3).
This basic framework has been developed to include options for variably connected and
weighted spatial representations, with the potential to accurately represent processes in stream
networks (Dray et al. 2006). Another set of spatial eigenvectors were created with the Moran’s
eigenvector map (MEM) method using a binary connectivity matrix and a weighting matrix (stream
distance between sites) (Dray et al. 2006). This allowed variable connectivity and rate of transference
between sites to be taken into account in the construction of this metric (MEM, Table 3). A weighted
metric was constructed by multiplying a vector of weights to the table of connectance used in the MEM
(Dray et al. 2006). I used average discharge to weight the links between sites to more accurately
represent the rate at which solutes and particles in the water are transferred throughout the system
(MEM-W, Table 3).
Chapter Four 109
Table 3. Four different spatial metrics were tested on sites in the Cam River. Eigenfunction-based spatial filtering
techniques were used, allowing flexibility of weighting and connectivity in spatial representation. I used principal
coordinates of neighbour matrices (PCNM) with two distance metrics, and stream distance for the other spatial metrics. I
used symmetrical distance between flow connected sites, based on Moran’s eigenvector maps (MEM) and a corresponding
weighted metric (MEM-W, based on the average velocity encountered between sites) to represent different spatial
structuring processes.
Distance measurement Weight Code DescriptionEuclidean distance PCNM-E PCNM on Euclidean distances between
sites Stream distance PCNM-S PCNM on stream distances between sites
Direction – flow connectance
MEM MEM all sites directly connected by flow in both directions
Average discharge
MEM-weighted
MEM – weighted by average discharge encountered between sites.
The spatial metric that best described variation in ecosystem processes was chosen using the adjusted
coefficient of multiple determination (R2a) to compare the variance explained by each spatial metric
(Peres-Neto et al. 2006, Blanchet et al. 2008a). Each set of eigenvectors from each of the eigenfunction
analyses for each distance measure, was subjected to forward selection (α < 0.1) to detect eigenvectors
explaining the most variance in community structure (Blanchet et al. 2008a). The suite of ecosystem
processes, then each individually, were analysed as functions of the set of PCNM, MEM and AEM
eigenvectors by canonical redundancy analysis (RDA), a multivariate regression based analysis using
the spatial vectors as predictors for the ecosystem processes. The forward selected spatial variables that
explained the most variation were further used in variance partitioning.
I assessed the relative contribution of environmental processes versus other spatial processes on
multi-scale patterns in ecosystem processes using partial redundancy analysis (pRDA (Cottenie 2005,
Dray et al. 2006, Peres-Neto and Legendre 2010). The set of environmental variables used in each
analysis was chosen by forward selection in relation to the response variable(s) in question (as for the
spatial variables, Blanchet et al. 2008a). I attempted to isolate the spatial and environmental
contributions to the rates of ecosystem processes by including not only the environmental variables that
impacted each site locally (Table 1), but also a range of variables which defined the condition of the
Chapter Four 110
catchment feeding water to that site, and that of the riparian zone along upstream reaches (Table 2).
Sets of three matrices were used in pRDA, a spatial matrix (forward selected spatial variables from the
best spatial model) and environmental matrix and a response matrix (ecosystem processes or a single
variable for individual tests), to compare proportions of variation in the community explained by each
predictor variable matrix. The relative weight of each fraction of partitioned variance was represented
by the adjusted R2 (R2a) which controls for the number of sites (differing for chlorophyll), the number
of explanatory variables and the probability of detecting effects (Peres-Neto et al. 2006). The forward
selected environmental and spatial variables were first tested as non-saturated global models,
individually ([S] and [E] fractions). For each analysis, if either the global spatial model or global
environmental model was significant, I proceeded with the partitioning approach to test the relative
proportions of their independent effects on community structure (independent spatial [S|E] and
independent environmental [E|S] fractions) and finally the joint variation shared by space and
environment (Laliberté et al. 2008).
These analyses were conducted in the R-language environment (R Development Core Team
2007) using the packages “vegan” (Oksanen et al. 2007) for RDA and variation partitioning; “PCMN”
(Dray et al. 2006) for the construction of PCNM variables; and “packfor” for the selection of
explanatory variables in the RDA. In all tests of significance, 999 permutations were used. Following
Anderson and Legendre (1999) permutation of raw data is adequate for ANOVA as there are no outlier
values in the factors.
Chapter Four 111
4.3 RESULTS
4.3.1 Patterns in ecosystem processes
The abundance of chlorophyll a was significantly positively correlated with the reduction in organic
matter in leaf packs (rs = 0.369, p = 0.009), but no significant correlations occurred between the other
measures of ecosystem processes. Algal productivity was highest at sites in and around Rangiora, and
in main stems directly downstream of the town (Figure 2a); correspondingly the reduction in leaf mass
was also higher at main-stem sites downstream of Rangiora, compared with those above the town and
in the Coldstream branch (Figure 2b). Microbial respiration and the DIN:SRP ratio became lower
downstream, with various patches of lower levels punctuating the generally high levels in the upper
system (Figure 2c and d).
Chapter Four 112
Figure 2 – Maps showing the spatial variation in each measure of ecosystem processes; chlorophyll a concentration (algal
productivity) (a), reduction in organic mass of leaf packs (b), the rate of microbial respiration on those leaves (c) and the
DIN:SRP ratio (d), throughout the Cam River. The stream drains to the bottom right. The size of each circle represents the
magnitude of that parameter at that site (see individual keys for actual levels), the divisions were created using the “Jenks”
natural breaks function in ArcMap 9.2. The scale bar and North arrow apply to all maps.
There was a trend in all four measures of ecosystem processes towards less variation as average
discharge increased (Figure 3a-d). The relationship between organic matter loss in leaf packs and
discharge had a significant positive floor and ceiling, although only the floor is shown, as this is where
the limit relationship appears most observable (p < 0.001, Figure 3a, Table 4). The relationship
between chlorophyll a and average discharge also had a significant positive floor limit response (p =
0.015, Figure 3d, Table 4). Alternatively, the relationships of DIN:SRP and microbial respiration with
discharge has a negatively sloped upper ceiling (c and b). The response of these ecosystem processes to
stream size measures was only weakly significant (α < 0.1).
Chapter Four 113
Average discharge (m3/s)
0.0 0.5 1.0
Mic
robi
al re
spira
tion
(mg
of O
2 /
hr /
g of
leaf
)
0
2
4
6
8
10
12
Average discharge (m3/s)
0.00 0.25 0.50 0.75
Chl
orop
hyll
a (m
g/m
2 )
0
25
50
75
100
125
150
175
0.0 0.5 1.0
Leaf
dec
ompo
sitio
n(g
lost
from
orig
inal
5g)
2.0
2.5
3.0
3.5
4.0
4.5
5.0
0.0 0.5 1.0
Rat
io D
IN:S
RP
0
250
500
750
1000
(a) (b)
(c) (d)
Figure 3 – Quantile regression relationships between average discharge and algal productivity (as chlorophyll a) (a), leaf
decomposition (reduction in organic matter in leaf packs) (b), microbial respiration on this leaf matter (c) and the DIN:SRP
ratio (d). The lower quantiles (10th) in (a) and (b) were significant at α < 0.05 and the upper quantiles (90th) in (b) and (c)
were significant at (α < 0.1). Note that Chlorophyll was measured at a subset of sites, thus the discharge axis is shorter.
Table 4 - Results of quantile regression, where discharge was the predictor variable for each response ecosystem process
variable. Values highlighted in bold were significant (α < 0.05) when assessed using bootstrapping methods.
4 Prop sheep-beef area -0.70 DIN 0.47 Prop arable area 0.36 Shade upstream 0.33
DIN:SRP ratio MEM 43
8 DIN 0.80 Prop sheep area 0.45 Conductivity 0.41 Site macro cover -0.14 Prop arable area 0.13 Prop lifestyle 100-m 0.13 Dairy area -0.035 Vegetation 100-m -0.03
Chapter Four 117Table 6. Spatial (S), environmental (E), independent spatial (S|E) and independent environmental (E|S) partitions of variation, and joint variation in ecosystem
processes throughout the Cam River compared using partial redundancy analysis (pRDA). R2a, the adjusted correlation coefficient, was used to partition the variance
into four fractions (S|E, E|S, joint and residual) in an unsaturated model (see methods text for details). Values significant at † p ≤ 0.1, * p ≤ 0.05, ** p ≤ 0.01. ‘Joint’
variation is the component shared by spatial and environmental variation which cannot be statistically separated.
Figure 4: The relative contribution of the independent spatial (S|E), independent environmental (E|S) partitions of variation, joint variation and unexplained (residual)
variation in ecosystem processes; as a combined measure (a), algal productivity (b), leaf decomposition (c), microbial respiration (d) and the DIN:SRP ratio (e),
throughout the Cam River. This was assessed using partial redundancy analysis (pRDA). The adjusted correlation coefficient, R2a, was to calculate the weight of each
fraction (size of the pie slice, see methods text for further details). The significance of each fraction, in each canonical pRDA model is indicated by the p value
superimposed on the corresponding pie segment; † ≤ 0.1, * P ≤ 0.05, ** P ≤ 0.01. The E|S fractions of (b) and (e) were not significant (p values not shown). Note that
the joint and unexplained fractions cannot be tested for significance.
Chapter Four 118
(a) Combined ecosystem processes
S|EE|S JointUnexplained
p = 0.051†
p = 0.005**
(a) Algal productivity
p = 0.005**
p = 0.005**
p = 0.005**
(c) Leaf decomposition
p = 0.005**
p = 0.04*
p = 0.005**
(d) Microbial respiration
p = 0.005**
p = 0.06†
(e) DIN:SRP ratio
p = 0.005**
Chapter Four 119
4.4 DISCUSSION
The spatial pattern of ecosystem processes varied in my study depending on which component was
under observation and in a way that was influenced by stream size. All spatial patterns resulted in
increased heterogeneity of ecosystem processes in headwaters compared with mainstems. The
directional connectivity and spatial separation of sites on the stream network interacted with
environmental factors resulting in variable ecosystem processes throughout the Cam River. I used
recently developed spatially explicit variance partitioning methods (Dray et al. 2006, Peres-Neto and
Legendre 2010) to tease apart the relative importance of the spatial and environmental structure of
the Cam River in determining the spatial patterns of each ecosystem process.
4.4.1 Patterns in ecosystem processes in the Cam River
Although the rates of algal productivity, leaf decomposition and microbial respiration on leaves in
streams often have similar responses to variation in nutrient levels and temperature (Biggs and
Kilroy 2004, Young et al. 2008), the only measures positively correlated with each other were algal
productivity and leaf matter decomposition. This indicates that these two processes may be
responding to similar stressors at similar scales. A correlation was also expected between the
DIN:SRP ratio and each of the these ecosystem process measured, due to a strong link between this
ratio and nutrient levels. However, the lack of a correlation in the Cam River is not surprising
considering the inconsistent relationship between algal productivity (Biggs and Kilroy 2004), leaf
decomposition (Young et al. 2008) and nutrients, across described studies. The lack of correlation
between leaf decomposition and microbial respiration is unusual however, considering the key role
the microbes often have in organic matter breakdown in streams, as well as the findings of previous
studies (Young et al. 2008).
Chapter Four 120
The high levels of algal productivity and leaf matter decomposition at sites in Rangiora and
along the mainstems of the North and Southbrooks are in keeping with the positive correlation
relationship I found between these measures. Lower levels of algal production occurred in upper
Coldstream branch where shading was considerable, and in the upper reaches of the North and
Southbrooks above Rangiora where flow was largely from water-races that connect with the system
at these points. The amount of chlorophyll a measured at almost all sites in the Cam River network
falls within the range associated with very impacted or eutrophic streams (25-260 mg-1m-2, mean <
70 mg-1m-2, Dodds et al. 1998). The method of assessing chlorophyll a I used was different from the
sampling of benthic stones suggested by Dodds et al. (1998), which would have some level of
grazing by invertebrates. Although my methods are not directly comparable to these guidelines, they
are likely to be a better reflection of eutrophication, as they are independent of variation in grazing
that may be site-specific. As the amount of chlorophyll I measured was often close to or exceeding
70 mg-1m-2, these were clearly extremely productive streams with very high algal abundance.
Although high levels of algal production are often used as an indicator of pollution in
streams (Fellows et al. 2006), benthic algae provide the main source of energy for higher trophic
levels in many unshaded temperate streams and their uptake of nutrients can help purify the streams
(Biggs 1996). In this respect, invertebrates inhabiting reaches of the Cam River with high algal
concentrations may benefit from this abundant resource. However, in parts of the Cam River, algae
were so abundant that thick mats had formed, clogging the stream and very likely being difficult for
the majority of invertebrates to graze because of their filamentous nature. The thick algal mats
meant that many sites were heavily silted and likely had poorly functioning biotic communities as a
result (Lowe and Laliberté 2008). Previous studies have found algal primary production rates to vary
between the major different habitats of streams at the reach scale and that this had an impact on
community study in each habitat (Keithan and Lowe 1985). Thus I would expect the large scale
Chapter Four 121
differences in algal productivity in the Cam River may have flow-on effects to the distribution of
aquatic organisms.
The rate of leaf litter decomposition at all sites on the Cam River was above the range
indicative of good ecosystem health (0.01 to 0.03g-1d-1 Gessner and Chauvet 2002). Based on these
guidelines the entire Cam River network has mild or severe impairment of ecosystem health. The
rate of organic matter decomposition throughout the Cam River was much higher than necessary to
maintain a healthy ecosystem (Gessner and Chauvet 2002). Terrestrial inputs, such as leaves, are an
important basal food resource supporting higher trophic levels (Benfield 2006). The rapid
decomposition process may mean that soluble inorganic forms of nutrients are being released into
the stream, rather than incorporated into the food web as organic form.
The high level of algal production and leaf litter decomposition I observed in the mainstems
of the Cam River is also shown in the positively sloped lower limit relationship with discharge, and
the forward selection of stream order, (as one of the group of environmental variables) which
explained a significant amount of spatial variation in these ecosystem processes in the global models
and when using partial redundancy analysis incorporating spatial structure (for decomposition only).
In these models, stream order was also positively correlated with algal production. These
relationships imply that larger streams are likely to have higher algal cover than many small
streams. This may be because larger reaches in the Cam River experience a greater flux of total and
dissolved phosphorus as well as dissolved nitrate (Chapter 3) and are less shaded than smaller
streams; all of which would stimulate cell growth and colony size.
The rate of microbial respiration on leaves and the DIN:SRP ratio became lower
downstream through the Cam River network. In contrast to algal productivity and leaf
decomposition rate, these two functional metrics displayed a negative upper limit relationship with
discharge. This results in the lack of correlation between leaf decomposition and microbial
respiration and indicates that sites on large reaches are less likely to have high rates of microbial
Chapter Four 122
decomposition. The mismatch between patterns in microbial respiration and leaf litter
decomposition suggests microbial respiration is not primarily responsible for leaf litter
decomposition. Other factors, such as the density of shredding invertebrates (Baldy et al. 1995,
Jonsson et al. 2001) and high flow velocity (Casas 1996) also increase breakdown rates and these
factors might have a greater influence on leaf litter breakdown than the action of bacteria and fungi
in my streams. The processes of microbial respiration and leaf decomposition are both likely to
respond to nutrient levels however, and in the Cam River it appears that each responds in a different
way.
The DIN:SRP ratio was much higher than the optimal ratio suggested as indicating healthy
amounts of algal production (16N: 1P, Redfield 1958), at almost all sites in the Cam River. A high
DIN:SRP ratio indicates that this system is largely phosphorus limited. However, while nutrient
ratios suggest potential phosphorus limitation, absolute nutrient concentrations must also be
considered. At levels above a certain threshold, limitation no longer occurs (Bothwell 1985). This
may be the case in the Cam River which has relatively high levels of both DIN and SRP (Chapter 1),
and may explain why algal productivity did not correlate with DIN:SRP ratio, and also why no
nutrients were selected as significant in predicting variation in algal production under forward
selection in my global models.
I found that variability in all measures of ecosystem functioning was higher in smaller
streams than in large streams. This is likely due to the higher spatial and temporal variability of
water quality in small streams (Chapter 2 and 3). The slope of the Quantile regression lines that
defined the limit relationships with stream size and microbial decomposition or DIN:SRP was less
steep than those of algal productivity and the rate of leaf decomposition, suggesting that stream size
had less influence over the former variables. Modification of riparian zones, be it through
agricultural or urban influences, results in increased stream temperature and nutrient levels, both of
which stimulate productivity (Young et al. 2008). However, riparian modification can also cause
Chapter Four 123
sedimentation (Gulis and Suberkropp 2003) and low concentrations of dissolved oxygen (Pascoal
and Cássio 2004), which can have negative impacts on algal and microbial activity. The extremes of
each physico-chemical parameter are experienced at smaller sites on the Cam River (Chapter 1 and
2). The variable response of productivity to these extremes is likely to result in variable types of
biotic communities across the headwaters of the Cam River.
4.4.2 How spatially structured were ecosystem processes?
Connectivity between sites was important in structuring the spatial pattern of ecosystem processes.
Spatial eigenvectors constructed using the MEM method, which incorporated the in-stream distance
and the connectivity pattern between sites, consistently explained the most spatial variation,
compared with the other methods tested. This suggested that the flow of solutes, organic matter,
sediments and propagules of algae or microbes between sites were important in determining the
spatial patterns in each measure of ecosystem processes.
Some spatial variation in all measures of ecosystem processes was modelled by eigenvectors
with small values which describe local structures (Borcard and Legendre 2002). This is likely due to
the influence of sites directly connected by flow, which are thus under very similar water quality
conditions and potentially the same land-use, riparian and in-stream conditions. However, with all
ecosystem process measures, significant spatial vectors also corresponded to the larger scales which
describe global structures (Borcard and Legendre 2002). This indicated that the distance and
upstream-downstream flow connectivity between sites explained variability in ecosystem processes
at multiple spatial scales. The gap in the range of spatial predictors at the small to medium scale may
indicate that the entry of tributaries disrupts the spatial pattern. At larger scales, the impact of
confluences is not felt, potentially buffered-out by the volume of water in the main stem, or its
overriding landscape or in-stream conditions. The composite measure of ecosystem processes is
spatially structured across all scales, as it takes into account the spatial structuring of each type of
Chapter Four 124
ecosystem process measured, which have different absolute spatial patterns and levels of
autocorrelation. Composite measures of ecosystem processes are more likely to reflect broad scale
spatial and environmental influences, and should be considered for use when attempting to manage
system wide stream health.
4.4.3 Spatial versus environmental control of ecosystem processes
The variation partitioning technique I used in this study has been used to identify the various
combinations of spatial and environmental controls on community structure and then related to the
metacommunity processes that contribute to variation in species assemblages (Cottenie 2005, Peres-
Neto and Legendre 2010). These processes can be simplified into a gradient between: neutral or
patch dynamics models (communities structured primarily by species loss or dispersal) which can be
inferred when spatial structure independent of environmental structure is detected; and niche models
(communities structured primarily by the local environment) which can be inferred when
communities and environmental structure independent of spatial structures (Leibold et al. 2004,
Cottenie 2005). The flexibility of this method means that it can be easily transferred to streams,
where the spatial portion can be used to describe processes related to downstream dispersal of
organisms (Brind-Amour 2005) and water more generally (Lacey et al. 2007).
I have taken a novel approach and extended the application of variation partitioning to the
task of detecting the processes defining variation in ecosystem processes throughout a stream
network. Each of the ecosystem processes I investigated is likely to be related to the condition of the
water passing through a reach, which through the flowing nature of streams is highly spatially
structured, particularly in the Cam River (Chapter 1). However, the processes are also likely to be
governed by the in-stream, riparian and land-use conditions at each site, and thus will also be
environmentally structured. I attempted to isolate the spatial and environmental contributions to the
rates of ecosystem processes by defining the environmental effects at a site, not only locally, but
Chapter Four 125
also by the condition of the environment and the catchment feeding water to that site, and by the
condition of the riparian zone along upstream reaches. In this way, spatial effects of water flowing
downstream are isolated from the environmental effects that may be conditioning the water on its
path downstream to a particular site. The forward selection procedure I used to select significant
environmental variables, from the many I defined, took into account the level of inter-correlation
between variables when selecting those that best explained the spatial pattern in each measure of
ecosystem processes.
All processes studied were highly spatially structured, but the environment (with the
exception of the rate of leaf decomposition) and the relative role of space and the environment also
varied independently. The large independent effect of spatial structure on algal production,
microbial respiration and the DIN:SRP ratio suggests that dispersal of nutrients or water of
particular physical form (i.e. high temperature, or turbid) from upstream sites is an important
controller of spatial variability in these ecosystem processes. For algae, the downstream dispersal of
colonising propagules is also likely to be an important factor in determining the spatial patterns of
algal productivity (Swanson and Bachmann 1976). The composite measure of ecosystem processes
is also spatially structured, thus the water quality of the many small tributaries of the Cam River has
a large impact on the rates of ecosystems processes at down-stream sites, and in turn, the health of
sites for the biota that inhabit them.
Space had no independent impact on leaf decomposition rates throughout the Cam River,
implying that local conditions are most important for this variable. This is confirmed by the largest
portion of variance explained by environmental factors, being for leaf decomposition. Therefore,
variation in leaf decomposition rates may be useful for detecting local impacts when this is the goal
of a study, yet less useful for system wide monitoring (particularly so when few sites are used).
There was a large shared component of variation in each measure of ecosystem processes, where
spatial structuring and environmental processes were confounded. In these situations, either or both
Chapter Four 126
could be important but they cannot be statistically separated. The largest amount of unexplained
variation was present in the model of spatial and environmental impacts on the composite measure.
As much variation in the composite measure of ecosystem processes could not be explained by
space or the environment, this may not be the most useful way to study the impacts of land-use
intensification and riparian loss on stream health.
The specific environment of each site had a significant influence on microbial respiration,
leaf matter decomposition, and the composite measure of ecosystem processes. The environmental
fraction was greatest for leaf decomposition. Although these two measures were significantly
influenced by the environment, different variables were important within this environmental
contribution. This difference in environmental drivers of spatial patterns helps to explain why leaf
decomposition and microbial respiration were not correlated as I expected. Size-related variables,
stream order and upstream natural stream length are important for leaf matter decomposition, which
decreased with increasing stream size (as shown in the Quantile regression); whereas nitrate, and
proportion of a catchment in sheep and beef, were important for microbial respiration. As expected,
nitrate had a positive relationship with microbial respiration, comparable to the findings of others
(Fuss and Smock 1996). The influence of sheep and beef is negative; this may be driven by
increased sedimentation caused by stock access to waterways, which has been shown to reduce
respiration rates (Hagen et al. 2006, Gulis and Suberkropp 2003). As changes in land-use alter
multiple conditions that impact on metabolic rates, many studies reveal, similarly to mine, that
multiple interacting factors are responsible for respiration on leaf litter (Niyogi et al. 2003).
Stream order and nitrate were selected for across most models. Although the environmental
fractions they contributed to were not always significant, this highlights their influence over
ecosystem processes in general. Both were selected as explaining variation in the composite
measure of processes, although the environmental fraction (when combined with space) only
explained 6% of the variation in the composite measure - this was still significant. A small but
Chapter Four 127
significant link between a stressor such as nitrate and stream health is important, especially in light
of the high level of spatial structuring in the Cam River. This information could allow managers to
target particular locations in Cam River where nitrate input to the stream should be monitored,
especially as much nitrate is contributed by ground water upwelling in this system (Chapter 2). This
is another example of the applicability of a spatially explicit approach to targeted within-catchment
management.
My adaption of the variance partitioning approach (originally designed for use in community
ecology) to identify factors controlling ecosystem processes has proven the flexibility of such
methods and should be an encouragement to others in many fields of science to consider their use,
especially in situations where spatial correlation exists. One caveat, however, is that care that must
be taken with this approach, as the results depend on the quality and relevance of environmental data
available, and conclusions must take this into account (Laliberté et al. 2008). It may also
overestimate spatial influence, but at least then we can conclude strongly for any environmental
effects we find, compared with other studies that make conclusions despite not considering spatial
contributions.
The flexibility of spatial eigenfunctions methods has made them applicable to modelling
spatial variation in streams. By incorporating such a spatially explicit approach in my analysis, I
have been able to successfully tease apart factors driving the processes central to the functioning of
streams. In such a complex system as the Cam River it would be difficult and laborious to examine
the spatial patterns from a purely observational approach. Overall, my work highlights the
importance of protecting headwaters as these are where the extremes occur. Also, due to spatial
connectivity within systems such as the Cam River, the water quality and conditions of the
headwaters propagate downstream. Rates of ecosystem processes are a more readily understood
measure of stream health than water chemistry alone, and may be line with specific end-targets of
management, such as to reduce the level of algal from a visual perspective.
References 128
5 REFERENCES
Allan J. D. 2004. Landscapes and riverscapes: The influence of land use on stream ecosystems.
Annual Review of Ecology, Evolution, and Systematics 35:257-284.
American Public Health Association (APHA). 1995. Standard methods for the examination of water
and wastewater. American Public Health Association, American Water Works Association,
and Water Pollution Control Federation. 19th edition, Washington, DC.
Anderson, M. J. and P. Legendre. 1999. An empirical comparison of permutation methods for tests
of partial regression coefficients in a linear model. Journal of Statistical Computation and
Simulation 62:271-303.
Baker, M. E., D. E. Weller, and T. E. Jordan. 2007. Effects of stream map resolution on measures of
riparian buffer distribution and nutrient retention potential. Landscape Ecology 22:973-992.
Baldy, V., M. Gessner, and E. Chauvet. 1995. Bacteria, fungi and the breakdown of leaf litter in a
large river. Oikos 74:93-102.
Barbour, M., J. Gerritsen, B. Snyder, and J. Stribling. 1999. Rapid bioassessment protocols for use
in streams and wadeable rivers. Environmental Protection Agency, Office of Water,
EPA/841/B-99-002, Washington, DC.
Barrodale, I. and F. Roberts. 1974. Solution of an overdetermined system of equations in the l1
norm. Communications of the ACM 17:319–320.
Bartkow, M. and J. Udy. 2004. Quantifying potential nitrogen removal by denitrification in stream
sediments at a regional scale. Marine and Freshwater Research 55:309-315.
Beale, C., J. Lennon, J. Yearsley, M. Brewer, and D. Elston. 2010. Regression analysis of spatial
data. Ecology letters 13:246-264.
Beecher, H., E. Dott, and R. Fernau. 1988. Fish species richness and stream order in Washington
State streams. Environmental Biology of Fishes 22:193-209.
Behrendt, H. and D. Opitz. 1999. Retention of nutrients in river systems: dependence on specific
runoff and hydraulic load. Hydrobiologia 410:111-122.
Bell, T., R. Freckleton, and O. Lewis. 2006. Plant pathogens drive density-dependent seedling
mortality in a tropical tree. Ecology letters 9:569-574.
References 129
Benda, L., N. Poff, D. Miller, T. Dunne, G. Reeves, G. Pess, and M. Pollock. 2004. The network
dynamics hypothesis: how channel networks structure riverine habitats. BioScience 54:413-
427.
Benfield, E. F. 2006. Leaf breakdown in stream ecosystems. pp. 579–590 in F. R. Hauer and G. A.
Lamberti (eds.). Methods in stream ecology. Academic Press, San Diego, CA.
Bernhardt, E., G. Likens, D. Buso, and C. Driscoll. 2003. In-stream uptake dampens effects of major
forest disturbance on watershed nitrogen export. Proceedings of the National Academy of
Sciences of the United States of America 100:10304.
Biggs, B. J. 1985. Algae - a blooming nuisance in rivers. Soil and Water 21:27-31.
Biggs, B. J. F. 1989. Biomonitoring of organic pollution using periphyton, South Branch,
Canterbury, New Zealand. New Zealand Journal of Marine and Freshwater Research 23:263-
274.
Biggs, B. J. F. 1995. The contribution of flood disturbance, catchment geology and land use to the
habitat template of periphyton in stream ecosystems. Freshwater biology 33:419-438.
Biggs, B. J. F. 1996. Patterns in benthic algae of streams. pp. 31-56 in R. J. Stevenson, M. L.
Bothwell and R. L. Lowe (eds.). Algal ecology: freshwater benthic ecosystems. Academic
Press, San Diego, CA.
Biggs B. J. F. and C. Kilroy. 2000. Stream Periphyton Monitoring Manual. National Institute of
Water and Atmospheric Research, Christchurch, New Zealand.
Biggs, B. J. F. and M. E. Close. 1989. Periphyton biomass dynamics in gravel bed rivers: the
relative effects of flows and nutrients. Freshwater biology 22:209-231.
Biggs, B. J. F., M. J. Duncan, I. G. Jowett, J. M. Quinn, C. W. Hickey, R. J. Davies-Colley, and M.
E. Close. 1990. Ecological characterisation, classification and modelling of New Zealand
rivers: an introduction and synthesis. New Zealand Journal of Marine and Freshwater
Research 24: 277-304.
Biggs, B. J. F. and C. Kilroy. 2004. Periphyton. pp 15.1-15.22 in J. M. Harding, P. Mosely, C.
Pearson and B. Sorrel, (eds.). Freshwaters of New Zealand. New Zealand Hydrological
Society and New Zealand Limnological Society, Christchurch, New Zealand.
Billen, G. and J. Garnier. 2000. Nitrogen transfers through the Seine drainage network: a budget
based on the application of theRiverstrahler'model. Hydrobiologia 410:139-150.
Bishop, K., I. Buffam, M. Erlandsson, J. Fölster, H. Laudon, J. Seibert, and J. Temnerud. 2008.
Aqua Incognita: the unknown headwaters. Hydrological Processes 22:1239-1242.
References 130
Blanchet, F. G., P. Legendre, and D. Borcard. 2008a. Forward selection of explanatory variables.
Ecology 89:2623-2632.
Blanchet, F. G., P. Legendre, and D. Borcard. 2008b. Modelling directional spatial processes in